IR Questions

Question #1: Which Word2Vec model predicts the surrounding context words given
a target word?

• LSA

• GloVe

• FastText

• Skip-gram Model

• Continuous Bag of Words (CBOW)

Correct Answer: Skip-gram Model

Question #2: In the context of Retrieval-Augmented Generation (RAG), what is the

role of the "retrieval" component?

• To generate new text based on a given prompt.

• To evaluate the quality of the generated text.

• To fine-tune the pre-trained language model for a specific task.

• To fetch relevant information from an external knowledge source based on

the input query.

• To store and manage the model's learned knowledge.

Correct Answer: To fetch relevant information from an external knowledge source

based on the input query.

Question #3: What does it mean for a web crawler to be "scalable"?

• It is designed to increase the crawl rate by adding more machines.

• It can accurately translate web pages into multiple languages.

• It is able to crawl only secure websites.

• It can analyze website content for sentiment.

1|Page
 • It can adapt to different internet protocols.

Correct Answer: It is designed to increase the crawl rate by adding more machines.

Question #4: Which component of an LSTM network is responsible for deciding

what information to discard from the cell state?

• Forget Gate

• Hidden State

• Cell State

• Input Gate

• Output Gate

Correct Answer: Forget Gate

Question #5: When performing a basic search using Boolean operators, field
codes, truncation (*), and wildcard (?) symbols, which of the following best
describes how a system handles implicit politeness in search operations?

• The system automatically includes synonyms and plurals to refine searches

without direct commands.

• The system requires users to specify field labels from drop-down lists for
search refinement.

• The system uses explicit politeness commands to adapt searches to user

intent.

• The system limits search results by date ranges and full-text availability
through a results screen tab.

• The system employs advanced methods like word association to improve

query relevance.

Correct Answer: The system uses explicit politeness commands to adapt searches
to user intent.

2|Page
Question #6: A crawler is designed to operate in a distributed environment. Which
of the following is the most important consideration for maintaining data
consistency across all crawler nodes?

• Requiring all nodes to access the web through a single proxy server.

• Implementing a centralized URL frontier with proper synchronization

mechanisms.

• Ensuring all nodes have identical hardware configurations.

• Ensuring all nodes use the same operating system.

• Distributing URLs randomly to each node.

Correct Answer: Implementing a centralized URL frontier with proper

synchronization mechanisms.

Question #7: In the context of search engine evaluation, what is a "test collection"?

• A list of common spelling errors used to improve query correction.

• A tool used to monitor the performance of a search engine in real-time.

• A collection of users who are asked to test the search engine and provide
feedback.

• A set of documents, queries, and relevance judgments used to evaluate

search engines.

• A collection of algorithms used to rank search results.

Correct Answer: A set of documents, queries, and relevance judgments used to

evaluate search engines.

Question #8: Which method does Google use to resolve URLs during crawling in
the URL frontier?

• Using relative paths to construct complete URLs.

3|Page
 • Ignoring broken or redirecting URLs.

• Appending full URLs directly from search results.

• Applying the Mercator scheme for URL resolution.

• Only processing URLs found in meta tags.

Correct Answer: Applying the Mercator scheme for URL resolution.

Question #9: When performing a basic search with a crawler, which of the following
is TRUE?

• Results are ordered by relevance first and then by chronological order.

• Search refinement options include limiting results to peer-reviewed journals

only.

• Wildcard (?) symbols are not supported in the crawler's search function.

• Truncation (*) can be used to shorten search terms during a basic search.

• Field codes must be specified for each search to limit term usage.

Correct Answer: Truncation (*) can be used to shorten search terms during a basic
search.

Question #10: Which of the following is a significant challenge associated with

using Large Language Models (LLMs)?

• LLMs are easy to train with small datasets.

• LLMs are limited to understanding only one language.

• LLMs require very little computational power to run.

• LLMs are unable to generate human-like text.

• LLMs often suffer from hallucinations and may generate factually incorrect
information.

4|Page
Correct Answer: LLMs often suffer from hallucinations and may generate factually
incorrect information.

Question #11: A web crawler needs to prioritize crawling pages that are most likely
to provide new and valuable information to users. Which of the following URL
prioritization strategies would be most effective?

• Breadth-first crawling, starting from a set of seed URLs.

• Using a combination of PageRank, topic relevance, and freshness estimates

to dynamically prioritize URLs in the frontier.

• Depth-first crawling, following links as deeply as possible within each site.

• Randomly selecting URLs from the frontier.

• Prioritizing URLs based on their length, with shorter URLs crawled first.

Correct Answer: Using a combination of PageRank, topic relevance, and freshness

estimates to dynamically prioritize URLs in the frontier.

Question #12: Which of the following tools or methods allow for advanced search
techniques such as using Boolean operators, field codes, truncation, and
wildcards?

• Truncation and wildcards only

• Field codes only

• Explicit politeness techniques

• Boolean operators, field codes, truncation, and wildcards

• Boolean operators only

Correct Answer: Boolean operators, field codes, truncation, and wildcards

Question #13: Which of these poses a challenge for web crawlers?

• Consistent website design.

5|Page
 • Standardized HTML code.

• High server uptime.

• Site mirrors and duplicate pages.

• Low network latency.

Correct Answer: Site mirrors and duplicate pages.

Question #14: A search engine's P@10 is 0.4. Which of the following interpretations
is correct?

• The search engine is performing poorly.

• The 4th relevant document appears at rank 10.

• 40% of all documents in the collection are relevant to the query.

• 40% of the top 10 documents retrieved are relevant to the query.

• The search engine retrieved 4 relevant documents.

Correct Answer: 40% of the top 10 documents retrieved are relevant to the query.

Question #15: Which of the following scenarios requires the most careful
consideration of both explicit and implicit politeness?

• Crawling a website that explicitly allows unrestricted crawling.

• Crawling a government website during off-peak hours.

• Crawling a large university website with no [Link] file.

• Crawling a website known to be a mirror of another site.

• Crawling a small personal blog with a clear [Link] file.

Correct Answer: Crawling a large university website with no [Link] file.

Question #16: Which of the following is the primary reason for evaluating search
engines?

6|Page
 • To assess the aesthetic appeal of the search engine's user interface.

• To measure how well the search engine satisfies user information needs.

• To count the number of lines of code in the search engine's software.

• To determine the cost-effectiveness of the search engine's infrastructure.

• To analyze the programming language used to develop the search engine.

Correct Answer: To measure how well the search engine satisfies user information
needs.

Question #17: What does Recall measure in the context of information retrieval?

• The cost of retrieving documents.

• The proportion of relevant documents that are retrieved.

• The average rank of relevant documents.

• The proportion of retrieved documents that are relevant.

• The time it takes to retrieve documents.

Correct Answer: The proportion of relevant documents that are retrieved.

Question #18: A web crawler is encountering an increasing number of spam pages.

Which combination of techniques would be most effective in identifying and
avoiding these pages?

• Requiring CAPTCHA completion for every page crawled.

• Relying solely on [Link] directives and URL blacklists.

• Ignoring all pages with excessive use of keywords or advertisements.

• To Crawl only websites with high domain authority and established

reputations.

• Using machine learning models trained on content features, link analysis,

and user feedback, combined with adaptive crawling strategies.

7|Page
Correct Answer: Using machine learning models trained on content features, link
analysis, and user feedback, combined with adaptive crawling strategies.

Question #19: What is the purpose of the F1-measure?

• To determine the number of documents in a collection.

• To calculate the cost of indexing documents.

• To assess the user interface design of a search engine.

• To provide a single score that balances Precision and Recall.

• To measure the speed of document retrieval.

Correct Answer: To provide a single score that balances Precision and Recall.

Question #20: When Google crawls a webpage and caches its content, how does
this affect the determination of page relevance for search queries?

• The crawler does not use cached content for ranking purposes.

• The cached content is only for backup access.

• Relevance is based solely on the live page.

• Cached pages are never considered in relevance.

• The cached content is used to judge relevance.

Correct Answer: The cached content is used to judge relevance.

Question #21: What is a "spider trap" designed to do?

• Simplify website navigation.

• Improve website security.

• Optimize website loading speed.

• Cause a web crawler to make an infinite number of requests or crash.

• Enhance user experience.

8|Page
Correct Answer: Cause a web crawler to make an infinite number of requests or
crash.

Question #22: Which factor is least relevant when considering whether a page is
malicious?

• The page's attempts to collect personal information without consent.

• Whether the page redirects users to unexpected websites.

• Whether the page attempts to install software without user consent.

• The page's loading speed.

• The presence of excessive pop-up advertisements.

Correct Answer: The page's loading speed.

Question #23: Which of the following is a characteristic of a well-designed web

crawler?

• It only crawls websites in a single language.

• It ignores [Link] to ensure complete coverage.

• It is designed to run on multiple distributed machines.

• It prioritizes crawling websites with the fewest links.

• It operates on a single machine to avoid network issues.

Correct Answer: It is designed to run on multiple distributed machines.

Question #24: When Google crawls a webpage and caches its content, how does
this affect the determination of page relevance for search queries?

• The crawler does not use cached content for ranking purposes.

• The cached content is only for backup access.

• Relevance is based solely on the live page.

9|Page
 • Cached pages are never considered in relevance.

• The cached content is used to judge relevance.

Correct Answer: The cached content is used to judge relevance.

Question #25: How does PubMed handle search terms entered without a specific
field label?

• The system allows the imposition of limits by date, gender, age, language, or
field label to refine search results.

• PubMed does not automatically map terms but directly searches for the
exact phrase or combination entered in the query box.

• PubMed uses an automatic term mapping feature to search for terms using
Medical Subject Headings (MeSH) and controlled vocabularies. When a term
is entered without a field label, the system starts by searching in the MeSH
table and then explores more specific categories.

• Term mapping can be bypassed if terms are entered as phrases,

hyphenated, or using truncation symbols.

• PubMed displays results in batches of 20 citations, with sorting options for

author, journal title, and publication date.

Correct Answer: PubMed uses an automatic term mapping feature to search for
terms using Medical Subject Headings (MeSH) and controlled vocabularies. When a
term is entered without a field label, the system starts by searching in the MeSH
table and then explores more specific categories.

Question #26: What is the purpose of the URL frontier in a web crawler?

• To store the content of crawled web pages.

• To prevent the crawler from accessing certain websites.

• To display the crawled web pages to users.

• To manage and prioritize URLs to be crawled.

10 | P a g e
 • To encrypt the data transmitted by the crawler.

Correct Answer: To manage and prioritize URLs to be crawled.

Question #27: What does "continuous operation" mean for a web crawler?

• Crawling websites only once.

• Continuously fetching fresh copies of previously fetched pages.

• Crawling only the top-level domain of each website.

• Crawling websites in alphabetical order.

• Crawling the web only during daytime hours.

Correct Answer: Continuously fetching fresh copies of previously fetched pages.

Question #28: A web crawler encounters a URL that dynamically generates an

infinite number of pages. Which of the following strategies would be most effective
in preventing this spider trap from crashing the crawler?

• Disabling JavaScript execution in the crawler.

• Increasing the crawler's memory allocation.

• Prioritizing the crawling of known high-quality websites.

• Implementing a URL length limit and a path depth limit.

• Ignoring all URLs with query parameters.

Correct Answer: Implementing a URL length limit and a path depth limit.

Question #29: How does Google handle non-ASCII characters in domain names?

• By converting them into ASCII using the Punycode system.

• By ignoring non-ASCII characters entirely.

• By truncating the domain name to fit ASCII standards.

• By appending a special prefix to indicate language.

11 | P a g e
 • By translating them into Chinese characters.

Correct Answer: By converting them into ASCII using the Punycode system.

Question #30: What is the primary reason for using multi-head attention in the
Transformer architecture?

• To prevent overfitting by regularizing the attention weights.

• To improve the model's ability to generalize to unseen data.

• To accelerate the training process by distributing the attention computation

across multiple GPUs.

• To reduce the computational complexity of the self-attention mechanism.

• To enable the model to capture different types of relationships and

dependencies in the input sequence.

Correct Answer: To enable the model to capture different types of relationships

and dependencies in the input sequence.

Question #31: Which of the following machine learning paradigms involves learning
from labeled training data to make predictions?

• Reinforcement Learning

• Supervised Learning

• Semi-Supervised Learning

• Unsupervised Learning

• Active Learning

Correct Answer: Supervised Learning

Question #32: Which component of the Transformer architecture is responsible for

injecting information about the position of tokens in the input sequence?

12 | P a g e
 • Self-Attention Mechanism

• Positional Encoding

• Multi-Head Attention

• Feed-Forward Neural Network

• Layer Normalization

Correct Answer: Positional Encoding

Question #33: Which of the following is a core component of the Transformer

architecture?

• Convolutional Neural Networks (CNNs)

• Self-attention mechanism

• Bag of Words (BoW) model

• Recurrent Neural Networks (RNNs)

• Support Vector Machines (SVMs)

Correct Answer: Self-attention mechanism

Question #34: What is the primary purpose of Retrieval-Augmented Generation

(RAG) in the context of Large Language Models (LLMs)?

• To enable LLMs to perform real-time translation of text.

• To reduce the computational cost of training LLMs.

• To allow LLMs to generate code in multiple programming languages.

• To improve the factual accuracy and reduce hallucinations in LLM-generated

text.

• To enhance the creativity and artistic style of LLM-generated content.

Correct Answer: To improve the factual accuracy and reduce hallucinations in

LLM-generated text.

13 | P a g e
Question #35: Level 3: Consider the query "A AND (B OR C)". Which processing
order is generally most efficient, assuming postings lists are sorted by DocID?

• Process "A AND C" first, then "OR" with B

• Process A, B, and C independently, then combine

• Process "A AND B" first, then "OR" with C

• It doesn't matter; all orders are equivalent

• Process "B OR C" first, then "AND" with A

Correct Answer: Process "B OR C" first, then "AND" with A

Question #36: What information is stored in a positional index (slide 7) that is NOT
stored in a non-positional index?

• A list of all terms in the dictionary

• The document frequency of each term

• The positions of each term's occurrences within each document

• The total number of documents in the collection

• The weight of each term in each document

Correct Answer: The positions of each term's occurrences within each document

Question #37: In the context of k-gram indexes for wildcard queries, what is the
purpose of the post-filtering step?**

• To remove terms that don't actually match the original wildcard query

• To sort the results by relevance

• To calculate the Jaccard coefficient

• To combine the results from multiple k-gram queries

• To create the k-gram index itself

14 | P a g e
Correct Answer: To remove terms that don't actually match the original wildcard
query

Question #38: Why might a term-document incidence matrix be impractical for

large document collections?

• It would be too dense, requiring excessive memory

• It cannot handle phrase queries

• It cannot be used for ranking

• It would be extremely sparse, wasting a lot of space

• It cannot handle Boolean queries

Correct Answer: It would be extremely sparse, wasting a lot of space

Question #39: In a term-document incidence matrix, what does a '1' in cell (i, j)
represent?

• Term i appears in document j

• Term i and document j are semantically related

• Document j is the i-th document in the collection

• Document j contains term i

• Term i is the j-th most frequent term in the collection

Correct Answer: Term i appears in document j

Question #40: Question: What is a "proximity operator" in the context of

information retrieval?

• An operator that ranks documents based on their relevance to a query

• An operator that specifies that two terms must occur close to each other in a
document

15 | P a g e
 • An operator that automatically corrects spelling errors in a query

• An operator that finds documents related to a specific topic

• An operator that translates a query into different languages

Correct Answer: An operator that specifies that two terms must occur close to
each other in a document

Question #41: What does the symbol '*' typically represent in a wildcard query?

• The end of a term

• Any single character

• A specific character defined elsewhere

• The beginning of a term

• Any (possibly empty) string of characters

Correct Answer: Any (possibly empty) string of characters

Question #42: Question: What is the purpose of next word index?

• To store the sentiment of the next sentence in the document

• To store the geographical location of the next document in the collection

• To expand the query with synonyms of the original query terms

• To store the citation information of the next article in the collection

• To store the next word that follows each term in a document

Correct Answer: To store the next word that follows each term in a document

Question #43: What is a potential drawback of the Extended Boolean Model

compared to ranked retrieval models?**

• It doesn't handle phrase queries

• It doesn't provide a ranking of results based on relevance

16 | P a g e
 • It can't handle proximity operators

• It only works with small collections

• It's more complex to implement

Correct Answer: It doesn't provide a ranking of results based on relevance

Question #44: Which of the following is a key difference between BERT and GPT in
terms of their architecture and training?

• GPT is pre-trained on a smaller dataset compared to BERT.

• BERT uses a decoder-only architecture, while GPT uses an encoder-only

architecture.

• BERT uses bidirectional context understanding, while GPT uses

unidirectional context.

• BERT is primarily used for text generation, while GPT is used for text
understanding.

• GPT relies on masked language modeling, while BERT relies on next

sentence prediction.

Correct Answer: BERT uses bidirectional context understanding, while GPT uses
unidirectional context.

Question #45: What is the primary objective of Masked Language Modeling (MLM)
in BERT's pre-training?

• Predict the next sentence in a sequence.

• Generate new sentences similar to the input.

• Predict randomly masked words in a sentence.

• Classify the sentiment of a sentence.

• Translate the sentence into another language.

Correct Answer: Predict randomly masked words in a sentence.

17 | P a g e
Question #46: Which of the following word embedding techniques leverages global
corpus statistics to create word representations?

• Word2Vec

• ELMo

• GloVe

• BERT

• FastText

Correct Answer: GloVe

Question #47: What is the primary goal of Word2Vec?

• To convert words into dense vectors that capture semantic relationships.

• To perform sentiment analysis on customer reviews.

• To identify the grammatical structure of sentences.

• To eliminate stop words from a text corpus.

• To translate text from one language to another.

Correct Answer: To convert words into dense vectors that capture semantic
relationships.

Question #48: What is the purpose of the Masked Language Modeling (MLM)
objective in BERT?

• To predict randomly masked words in a sentence, forcing the model to

understand bidirectional context.

• To classify the sentiment of a given text.

• To predict the next sentence in a sequence.

• To translate text from one language to another.

18 | P a g e
 • To generate new text based on a prompt.

Correct Answer: To predict randomly masked words in a sentence, forcing the

model to understand bidirectional context.

Question #49: What is the "scaling hypothesis" in the context of Large Language
Models (LLMs)?

• Increasing the batch size dramatically improves performance.

• Decreasing the learning rate dramatically improves performance.

• Increasing the amount of regularization dramatically improves performance.

• Decreasing the model size dramatically improves performance.

• Increasing model size and training data dramatically improves performance.

Correct Answer: Increasing model size and training data dramatically improves
performance.

Question #50: What is the primary reason that Large Language Models (LLMs) are
often paired with vector databases?

• To allow the LLM to perform mathematical calculations.

• To enable the LLM to access and retrieve relevant information from a large
corpus of knowledge.

• To reduce the size of the LLM.

• To improve the LLM's ability to generate code.

• To provide the LLM with a persistent memory of past interactions.

Correct Answer: To enable the LLM to access and retrieve relevant information
from a large corpus of knowledge.

Question #51: Which of the following is NOT a key component of the Transformer
architecture?

19 | P a g e
 • Self-Attention Mechanism

• Positional Encoding

• Multi-Head Attention

• Feed-Forward Neural Network

• Recurrent Layers

Correct Answer: Recurrent Layers

Question #52: Which of the following techniques is used to break down words into
subwords to handle out-of-vocabulary words in BERT?

• GloVe

• TF-IDF

• WordPiece tokenization

• Word2Vec

• Bag of Words

Correct Answer: WordPiece tokenization

Question #53: In Retrieval-Augmented Generation (RAG), what is the role of the

retrieval component?

• To evaluate the quality of the generated output.

• To fine-tune the language model for a specific task.

• To generate new text based on the input prompt.

• To compress the input prompt for faster processing.

• To extract relevant information from external knowledge sources.

Correct Answer: To extract relevant information from external knowledge sources.

20 | P a g e
Question #54: Which of the following best describes the function of the "forget
gate" in an LSTM (Long Short-Term Memory) network?

• It adds new information to the hidden state.

• It decides what information to discard from the cell state.

• It regulates the information passed to the next time step or the output layer.

• It determines what new information to store in the cell state.

• It combines the input gate and output gate functionalities.

Correct Answer: It decides what information to discard from the cell state.

Question #55: What is the primary purpose of prompt engineering in the context of
Large Language Models (LLMs)?

• To craft inputs that elicit desired behaviors from LLMs.

• To optimize the hardware infrastructure for LLMs.

• To fine-tune LLMs for specific downstream tasks.

• To reduce the computational cost of running LLMs.

• To encrypt sensitive data used to train LLMs.

Correct Answer: To craft inputs that elicit desired behaviors from LLMs.

Question #56: In Word2Vec, what is the main goal of the Skip-gram model?

• To cluster similar documents together.

• To predict the target word from its surrounding context.

• To predict the sentiment of a given text.

• To predict context words from a target word.

• To predict the next sentence in a sequence.

Correct Answer: To predict context words from a target word.

21 | P a g e
Question #57: In the context of Transformer architecture, what is the role of
positional encoding?

• To inject information about the position of tokens in the sequence.

• To normalize the input data.

• To capture semantic relationships between words.

• To mask certain tokens from the attention mechanism.

• To reduce the dimensionality of input embeddings.

Correct Answer: To inject information about the position of tokens in the sequence.

Question #58: Which model architecture is primarily decoder-based and focused

on text generation?

• GloVe

• RNN

• BERT

• GPT

• Word2Vec

Correct Answer: GPT

Question #59: What key innovation allowed Transformers to overcome the

limitations of RNNs in capturing long-range dependencies?

• Convolutional layers

• Attention Mechanism

• Pooling layers

• Activation functions

• Backpropagation

Correct Answer: Attention Mechanism

22 | P a g e
Question #60: Which of the following is a key advantage of Transformers over
Recurrent Neural Networks (RNNs)?

• Transformers can process entire sequences in parallel, allowing for faster

computation.

• RNNs utilize the self-attention mechanism.

• Transformers have difficulty capturing long-range dependencies.

• Transformers process data sequentially, while RNNs process data in parallel.

• RNNs are better at capturing global context.

Correct Answer: Transformers can process entire sequences in parallel, allowing

for faster computation.

Question #61: What is the main advantage of using LSTMs over basic RNNs for
processing sequential data?

• LSTMs require less computational power than RNNs.

• LSTMs can process data in parallel, while RNNs must process sequentially.

• LSTMs do not need to be trained on large datasets.

• LSTMs mitigate the vanishing gradient problem, allowing them to capture

long-range dependencies.

• LSTMs are simpler to implement.

Correct Answer: LSTMs mitigate the vanishing gradient problem, allowing them to
capture long-range dependencies.

Question #62: In the context of BERT, what is Masked Language Modeling (MLM)?

• A pre-training objective where the model predicts randomly masked words in

a sentence.

• A type of data augmentation used to increase the size of the training dataset.

23 | P a g e
 • A technique for generating new sentences from a given input.

• A fine-tuning technique for improving model performance on specific tasks.

• A method for predicting the next sentence in a sequence.

Correct Answer: A pre-training objective where the model predicts randomly

masked words in a sentence.

Question #63: What is a primary limitation of the Bag of Words (BoW) approach in
NLP?

• It reduces dimensionality effectively.

• It loses word order and treats text as an unordered collection of words.

• It captures semantic relationships between words.

• It effectively handles long-range dependencies.

• It preserves the grammatical structure of sentences.

Correct Answer: It loses word order and treats text as an unordered collection of
words.

Question #64: Which pre-training objective in BERT helps the model understand
relationships between sentences?

• Position Embeddings

• Transfer Learning

• WordPiece Tokenization

• Masked Language Modeling (MLM)

• Next Sentence Prediction (NSP)

Correct Answer: Next Sentence Prediction (NSP)

Question #65: When processing text, why might diacritics be removed?

24 | P a g e
 • To preserve the case sensitivity of proper nouns

• To improve search engine rankings

• Because they have minimal impact on query performance

• To handle acronyms correctly

• To standardize all words to lowercase

Correct Answer: Because they have minimal impact on query performance

Question #66: How does stemming generally affect retrieval performance in

English?

• It consistently improves recall but may harm precision.

• It consistently harms both precision and recall.

• It consistently improves precision but may harm recall.

• It consistently improves both precision and recall.

• It has no effect on either precision or recall.

Correct Answer: It consistently improves recall but may harm precision.

Question #67: Level 5: A user submits a complex query with many terms, some
with high document frequencies and some with low. What is a good strategy for
query processing?

• Process terms in order of decreasing document frequency.

• Process terms in order of increasing document frequency.

• Process all terms simultaneously.

• It depends on the specific retrieval model being used.

• Process terms randomly.

Correct Answer: Process terms in order of increasing document frequency.

25 | P a g e
Question #68: Which of the following BEST describes how the spelling correction
problem can be approached using heuristics?

• Inverted Index

• Term-Document Incidence Matrix

• Probabilistic Retrieval

• Boolean Retrieval

• Vector Space Model

Correct Answer: Boolean Retrieval

Question #69: In the Boolean retrieval model, how is a document viewed?

• A collection of sentences

• A set of words

• A hierarchical structure

• A sequence of characters

• A graph of terms

Correct Answer: A set of words

Question #70: What is an advantage of using a biword index?

• It is more accurate than a positional index

• It eliminates the need for stemming

• It can handle phrase queries directly

• It allows for efficient processing of single-word queries

• It reduces the size of the vocabulary

Correct Answer: It can handle phrase queries directly

26 | P a g e
Question #71: Question: What is a "posting" in the context of an inverted index?

• An entry in a postings list, recording that a term appeared in a document

• A web page that contains advertisements

• A statistical measure of a term's importance

• A user's query submitted to a search engine

• A document that is highly relevant to a query

Correct Answer: An entry in a postings list, recording that a term appeared in a

document

Question #72: Which of the following techniques is MOST directly related to

handling spelling variations in queries?

• Lemmatization

• Soundex

• n-gram indexing

• Stemming

• All of the above

Correct Answer: All of the above

Question #73: Why is it generally not advisable to treat an entire book as a single
document when performing searches?

• Makes searching faster

• Increases document size

• Reduces storage requirements

• Improves relevance by focusing on smaller sections

• Enhances encryption security

Correct Answer: Improves relevance by focusing on smaller sections

27 | P a g e
Question #74: Level 2: In Boolean retrieval, what is the difference between an
"information need" and a "query"?

• There is no difference.

• The query is broader, while the information need is more specific.

• The information need is expressed in natural language, while the query is in

Boolean logic.

• The information need is broader, while the query is the specific search string.

• The information need is processed by the system, while the query is not.

Correct Answer: The information need is broader, while the query is the specific
search string.

Question #75: When processing text to determine the vocabulary of terms, what is
a primary goal of lemmatization ?

• To convert words into their phonetic equivalents using Soundex.

• To remove all derivational affixes from a word, achieving proper

lemmatization.

• To identify and tag terms in multiple languages within a document

collection.

• To collapse derivationally related words into a single base form.

• To reduce inflectional endings to return the base lemma of a word.

Correct Answer: To reduce inflectional endings to return the base lemma of a

word.

Question #76: Why is logarithmic merging more efficient than repeatedly merging a
small auxiliary index into a large main index for dynamic indexing?

• Logarithmic merging distributes the indexing load across multiple machines

28 | P a g e
 • Logarithmic merging avoids sorting the main index

• Logarithmic merging reduces the number of disk seeks

• Logarithmic merging allows for parallel processing

• Logarithmic merging ensures that each posting is only merged a logarithmic

number of times

Correct Answer: Logarithmic merging ensures that each posting is only merged a
logarithmic number of times

Question #77: In query processing, what does "merging" postings lists typically
involve for an AND query?**

• Sorting the lists alphabetically

• Finding the intersection of the lists

• Concatenating the lists

• Finding the union of the lists

• Calculating the difference between the lists

Correct Answer: Finding the intersection of the lists

Question #78: In the Blocked Sort-Based Indexing (BSBI) algorithm, what is the
main reason for dividing the document collection into blocks?

• To reduce the number of disk seeks during sorting

• To allow for easier updates to the index

• To fit intermediate data into main memory

• To simplify the merging process

• To enable parallel processing on multiple machines

Correct Answer: To fit intermediate data into main memory

29 | P a g e
Question #79: In information retrieval, what does "edit distance" measure?

• The semantic similarity between two words

• The difference in length between two strings

• The minimum number of character-level operations (insert, delete,

substitute) to transform one string into another

• The number of documents containing both words

• The frequency difference between two words

Correct Answer: The minimum number of character-level operations (insert,

delete, substitute) to transform one string into another

Question #80: In Boolean Retrieval, what does the query "cat AND dog NOT mouse"
retrieve?

• Documents that contain "cat" and "dog" but not "mouse"

• Documents that contain "cat", "dog", or "mouse"

• Documents that contain both "cat" and "dog"

• Documents that contain "cat" or "dog" but not "mouse"

• Documents that contain "cat" and "dog" and may or may not contain
"mouse"

Correct Answer: Documents that contain "cat" and "dog" but not "mouse"

Question #81: Which of the following best describes the process of tokenization in
Information Retrieval?

• Creating equivalence classes of words, such as treating

• Breaking down a character sequence into individual semantic units.

• Reducing words to their root form, such as

• Grouping related documents together based on their content.

30 | P a g e
 • Identifying the language of a document.

Correct Answer: Breaking down a character sequence into individual semantic

units.

Question #82: Difficulty: level1 Question: Which of the following is NOT typically
considered a linguistic module operation in the indexing pipeline?

• Skip pointer creation

• Stemming

• Lemmatization

• Stop word removal

• Tokenization

Correct Answer: Skip pointer creation

Question #83: The algorithm described in Figure 2.12 is designed to find within-k
word proximity searches by using positional indexes. How does this approach
impact the required postings storage?

• They minimize storage requirements while supporting proximity searches.

• They do not affect the required postings storage.

• Positional indexes significantly increase required postings storage.

• They reduce the required postings storage.

• They allow for efficient computation of within-k proximity searches but may
trade off storage space.

Correct Answer: Positional indexes significantly increase required postings

storage.

Question #84: Skip pointers in postings lists are primarily used to:

31 | P a g e
 • Improve the efficiency of OR queries

• Handle phrase queries

• Improve the efficiency of AND queries

• Reduce the size of the postings lists

• Make the index dynamic

Correct Answer: Improve the efficiency of AND queries

Question #85: What symbol is commonly used to represent a wildcard in a

wildcard query?

• $

• %

• ?

Correct Answer: *

Question #86: A user frequently searches for information related to "automobile

repair," but sometimes uses "car repair" or "vehicle repair." Which combination of IR
techniques would BEST address this user's needs?

• Case folding and n-gram indexing.

• Stemming and stop word removal.

• Tokenization and stop word removal.

• Skip pointers and stemming.

• Lemmatization and thesaurus-based query expansion.

Correct Answer: Lemmatization and thesaurus-based query expansion.

32 | P a g e
Question #87: What is the purpose of a stop list in Information Retrieval?

• To store frequently occurring terms that are likely to be relevant to user

queries.

• To exclude common words like "the," "a," "is" from indexing.

• To store a list of all terms in the document collection.

• To maintain a list of synonyms for query expansion.

• To store a list of all stemmed words.

Correct Answer: To exclude common words like "the," "a," "is" from indexing.

Question #88: Which of the following is NOT a typical step in the process of
constructing an inverted index?

• Merging postings lists from different blocks (if using a blocked approach)

• Parsing documents to extract terms

• Calculating the semantic similarity between terms

• Creating postings lists for each term

• Sorting terms alphabetically

Correct Answer: Calculating the semantic similarity between terms

Question #89: How does a permuterm index handle a wildcard query like fi*mo*er?

• It computes the edit distance to all terms

• It only works for trailing wildcards

• It directly searches for the entire pattern

• It uses a k-gram index to find matching terms

• It breaks the query into smaller subqueries based on the * position, using
rotations.

33 | P a g e
Correct Answer: It breaks the query into smaller subqueries based on the *
position, using rotations.

Question #90: What is the primary purpose of the tokenization process in

Information Retrieval?

• To identify the language of a document

• To determine the document unit for indexing

• To chop a character sequence into pieces called tokens

• To perform word segmentation

• To convert the byte sequence of a document into a linear sequence of

characters

Correct Answer: To chop a character sequence into pieces called tokens

Question #91: In the context of Information Retrieval, what is an "inverted index"?

• A list of documents sorted by their relevance to a query.

• A compressed representation of the document collection.

• A list of all unique terms in the collection.

• A matrix showing the frequency of each term in each document.

• A data structure that maps terms to the documents they appear in.

Correct Answer: A data structure that maps terms to the documents they appear
in.

Question #92: What is the primary advantage of using a B-tree over a hashtable for
implementing a dictionary in an IR system?

• Lower memory consumption

• Simpler implementation

34 | P a g e
 • Faster lookup time

• Ability to handle prefix searches (wildcard queries)

• Easier handling of term variations

Correct Answer: Ability to handle prefix searches (wildcard queries)

Question #93: In a positional index, what information is stored in each posting, in

addition to the docID?

• A skip pointer

• The term frequency

• The stemmed form of the term

• The document length

• A list of positions (word offsets) where the term occurs in the document

Correct Answer: A list of positions (word offsets) where the term occurs in the
document

Question #94: Difficulty: 1 Question 2: In the Boolean Retrieval model (slide 7),
how is a document represented?

• As a graph of relationships

• As a sequence of characters

• As a weighted vector

• As a set of words

• As a probabilistic distribution

Correct Answer: As a set of words

Question #95: Question: In how many cases would a query of "o’neill" and "capital"
match?

35 | P a g e
 • 5

• 3

• 4

• 2

• 1

Correct Answer: 1

Question #96: What does edit distance allow in spelling correction techniques, as
described in the context?

• Measuring the number of differing characters between two strings.

• A combination of insertions, deletions, and replacements to find the

minimum steps between two strings.

• The minimum number of insertions or deletions needed to transform one

string into another.

• Correcting a single query term at a time.

• The number of replacements required to transform one string into another.

Correct Answer: A combination of insertions, deletions, and replacements to find

the minimum steps between two strings.

Question #97: Which of the following scenarios would be BEST addressed by using
a Soundex algorithm as part of the query processing pipeline?

• Expanding a query for 'car' to include 'automobile'.

• Correcting a query for 'acomodate' to 'accommodate'.

• Correcting a query for 'colour' to 'color'.

• Correcting a query for 'Britny Spers' to 'Britney Spears'.

• Handling a wildcard query like 'tele*one'.

36 | P a g e
Correct Answer: Correcting a query for 'Britny Spers' to 'Britney Spears'.

Question #98: What is the primary goal of an Inverted Index in Information

Retrieval?

• To allow for efficient calculation of the term-document incidence matrix.

• To enable grepping through text for efficient search.

• To map terms to the documents where they occur.

• To store the term-document incidence matrix in a compressed format.

• To store the complete text of each document for efficient retrieval.

Correct Answer: To map terms to the documents where they occur.

Question #99: A leading wildcard query is of the form ?

• mo*n

• m*n

• mon**

• mon*

• *mon

Correct Answer: *mon

Question #100: What is the purpose of query expansion?

• To automatically correct spelling errors in the query

• To add related terms to the query, potentially improving recall

• To remove stop words from the query

• To make the query shorter

• To restrict the query to a specific document collection

37 | P a g e
Correct Answer: To add related terms to the query, potentially improving recall

Question #101: Question: What is the primary purpose of an inverted index?

• To rank documents based on their relevance to a query

• To map terms to the documents in which they appear

• To provide a graphical user interface for search

• To store the original documents in a compressed format

• To perform stemming and lemmatization of terms

Correct Answer: To map terms to the documents in which they appear

Question #102: Which statement about the effectiveness of search engine indexing
methods is supported by the given context?

• Tomlinson's work (2003) suggested that lemmatizers performed worse than

stemming methods for most languages.

• Skip pointers, as discussed in Moffat and Zobel (1996), are no longer

considered effective for modern search engines due to performance issues.

• Using compound splitting improved search performance for all languages

discussed.

• Indexing based on character four-grams (n-grams) provided significant

improvements across multiple languages, including English and Spanish.

• The Bar-Ilan and Gutman study found that commercial web search engines
were hampered by a lack of language-specific processing in 2003.

Correct Answer: Indexing based on character four-grams (n-grams) provided

significant improvements across multiple languages, including English and
Spanish.

Question #103: What is a "stop word" in the context of IR?

38 | P a g e
 • A stemmed word

• A common word (e.g., "the," "a," "is") that is often removed during
preprocessing

• A rare word

• A word that appears in the query

• A misspelled word

Correct Answer: A common word (e.g., "the," "a," "is") that is often removed during
preprocessing

Question #104: What is the primary advantage of using skip pointers in postings
lists?

• They allow for faster merging of postings lists during query processing.

• They reduce the size of the index.

• They improve the accuracy of stemming.

• They help in identifying stop words.

• They eliminate the need for stemming.

Correct Answer: They allow for faster merging of postings lists during query
processing.

Question #105: Difficulty: level2 Question: What is the main difference between
stemming and lemmatization?

• Stemming is language-dependent, while lemmatization is not.

• Lemmatization always produces a valid word, while stemming may not.

• Stemming is only for English, while lemmatization works for all languages.

• Stemming is more computationally expensive than lemmatization.

• Lemmatization uses a dictionary, while stemming uses simple rules.

39 | P a g e
Correct Answer: Lemmatization uses a dictionary, while stemming uses simple
rules.

Question #106: Based on the exercise on slide 14, which of the following strategies
would be MOST efficient for processing the query (tangerine OR trees) AND
(marmalade OR skies) AND (kaleidoscope OR eyes)?

• Process the OR operations with the largest combined postings list sizes first.

• Process (tangerine OR trees) first, as it has the smallest individual term.

• Process the AND operations first, then the OR operations.

• Process the OR operations with the smallest combined postings list sizes
first.

• It doesn't matter what order the operations are processed in.

Correct Answer: Process the OR operations with the smallest combined postings
list sizes first.

Question #107: What type of query is a positional index essential for?

• Queries with stop words

• Boolean queries with AND

• Stemmed queries

• Phrase queries (e.g., "Stanford University")

• Boolean queries with OR

Correct Answer: Phrase queries (e.g., "Stanford University")

Question #108: A query for "flights to London" would likely benefit most from which
of the following IR techniques?

• Case folding

40 | P a g e
 • Keeping stop words

• Stop word removal

• Lemmatization

• Aggressive stemming

Correct Answer: Keeping stop words

Question #109: In soundex algorithm, the letters 'C', 'G', 'J', 'K', 'Q', 'S', 'X', 'Z' are
changed to what digit?

• 1

• 5

• 3

• 0

• 2

Correct Answer: 2

Question #110: What is the primary challenge in tokenization when dealing with
languages like Chinese and Japanese?

• The lack of punctuation marks.

• The absence of spaces between words.

• The large number of characters in the alphabets

• The use of right-to-left writing.

• The presence of multiple alphabets intermingled.

Correct Answer: The absence of spaces between words.

Question #111: A document collection contains many technical reports with

complex part numbers, error codes, and version numbers (e.g., "B-52",

41 | P a g e
"324a3df234cb23e"). Which indexing approach would be MOST helpful for
retrieving documents based on these specific identifiers?

• Lemmatization

• Stop word removal

• N-gram indexing

• Stemming

• Case folding

Correct Answer: N-gram indexing

Question #112: What is a "stop word" in Information Retrieval?

• A word that appears frequently in a document

• A word that is misspelled

• A word that is grammatically incorrect

• A very common word deemed unhelpful for retrieval, often excluded from
the vocabulary

• A word that is used in a phrase query

Correct Answer: A very common word deemed unhelpful for retrieval, often
excluded from the vocabulary

Question #113: Which of the following documents would NOT appear in the search
results for the query "Brutus and Caesar and not Calpurnia"?

• Antony and Cleopatra

• Romeo and Juliet

• The Tempest

• Hamlet

• Julius Caesar

42 | P a g e
Correct Answer: Hamlet

Question #114: What is an effective strategy used by some Boolean retrieval

systems like Westlaw and Lexis-Nexis to improve query generalization?

• Using compound splitter modules for German compound nouns.

• Handling language-specific reduced definite articles, such as the

apostrophe use in French (e.g., l’ensemble).

• Performing character-based indexing regardless of word boundaries.

• Encouraging users to enter hyphens in queries to allow generalization of

forms like "multi-word" to "multi word".

• Combining all of the above strategies for optimal indexing.

Correct Answer: Combining all of the above strategies for optimal indexing.

Question #115: An IR system is designed to handle queries in multiple languages.

Which of the following indexing strategies would be MOST robust and adaptable?

• Using language-specific tokenization and normalization modules.

• Ignoring all accents and diacritics.

• Using a single, universal stemming algorithm.

• Treating all words as case-insensitive.

• Relying solely on n-gram indexing.

Correct Answer: Using language-specific tokenization and normalization modules.

Question #116: What is the primary basis for ranked retrieval in information
retrieval systems?

• Boolean logic

• Index size

43 | P a g e
 • Term frequency

• Scoring functions

• Number of documents in the collection

Correct Answer: Scoring functions

Question #117: Which of the following is a characteristic of a "sparse" vector in the

context of the vector space model?

• It represents a common term.

• It is not used in information retrieval.

• It contains mostly non-zero values.

• It contains mostly zero values.

• It represents a short document.

Correct Answer: It contains mostly zero values.

Question #118: Consider a scenario where two different information retrieval

systems are being evaluated on the same dataset. System A achieves a higher
precision at k for low values of k (e.g., k=10) compared to System B. However,
System B achieves a higher MAP score. What inferences can be drawn about the
performance characteristics of these two systems, and how would you advise a
user who needs to choose between them?

• More information is needed to make a determination.

• The two systems are equivalent, and the choice between them is arbitrary.

• System B is better overall because MAP is a more reliable metric than

precision at k.

• System A is better overall because precision at k is a more reliable metric

than MAP.

44 | P a g e
 • System A is better for tasks where only the top few results matter, while
System B is better for tasks where finding as many relevant documents as
possible is important.

Correct Answer: System A is better for tasks where only the top few results matter,
while System B is better for tasks where finding as many relevant documents as
possible is important.

Question #119: Question: What is the purpose of using logarithmic frequency

weighting in term frequency calculation?

• To reduce the impact of very frequent terms.

• To convert term frequencies to binary values.

• To increase the impact of very frequent terms.

• To normalize term frequencies across different documents.

• To eliminate the need for inverse document frequency (IDF).

Correct Answer: To reduce the impact of very frequent terms.

Question #120: Which of the following factors is LEAST likely to affect the relevance
ranking of documents in a typical information retrieval system?

• Inverse document frequency.

• The color of the text in the document.

• The length of the document.

• Term frequency.

• The presence of query terms in the document title.

Correct Answer: The color of the text in the document.

Question #121: Question: In the vector space model, how are documents and
queries represented?

45 | P a g e
 • As vectors in a high-dimensional space.

• As linked lists of terms.

• As probability distributions over words.

• As binary matrices.

• As sets of keywords.

Correct Answer: As vectors in a high-dimensional space.

Question #122: How do Champion Lists relate to Index Elimination? Can they be
used together? Options:

• Champion Lists are implemented using a Term-Document Incidence Matrix

to track document relevance.

• Index Elimination is used without Champion Lists, relying solely on removing

lower quality documents.

• Champion Lists can be used with Index Elimination to improve search

results by focusing on high-quality documents.

• Index Elimination focuses on removing documents below a certain score

without considering Champion Lists.

• The use of Champion Lists requires the Inverted Index for effective
implementation.

Correct Answer: Champion Lists can be used with Index Elimination to improve
search results by focusing on high-quality documents.

Question #123: Question: What is the main purpose of using tf-idf weighting in
information retrieval?

• To reduce the computational complexity of query processing.

• To give equal weight to all terms in a document.

46 | P a g e
 • To assign higher weights to terms that are frequent in a specific document
but rare in the overall collection.

• To prioritize documents with the most terms.

• To rank documents based on their length.

Correct Answer: To assign higher weights to terms that are frequent in a specific
document but rare in the overall collection.

Question #124: Which of the following is true about inverted indexes in parametric
and zone indexes?

• Inverted indexes are used to store documents by term occurrences.

• Parametric indexes are not used for field-based queries.

• Zone indexes build an inverted index on specific document sections,

allowing efficient querying within those sections.

• All inverted indexes are designed solely for metadata fields.

• Parametric indexes are used for metadata fields like author or year.

Correct Answer: Zone indexes build an inverted index on specific document

sections, allowing efficient querying within those sections.

Question #125: Which of the following statements correctly describes cosine

similarity?

• After normalization, all documents have identical vectors regardless of their

original length and content.

• Cosine similarity is affected by the lengths of the document and query

vectors before normalization.

• Cosine similarity is calculated as the sum of the query weights and

document weights.

47 | P a g e
 • The cosine similarity formula subtracts the sum of squares from the dot
product to get the final score.

• When vectors are normalized by their L2 norm, cosine similarity becomes

equivalent to the dot product of the unit vectors.

Correct Answer: When vectors are normalized by their L2 norm, cosine similarity
becomes equivalent to the dot product of the unit vectors.

Question #126: Which of the following is a key limitation of using the Jaccard
coefficient for scoring in information retrieval?

• It cannot handle Boolean queries.

• It does not consider the frequency of terms within a document.

• It requires documents to be of equal length.

• It only works for short queries.

• It is computationally expensive to calculate for large document collections.

Correct Answer: It does not consider the frequency of terms within a document.

Question #127: In the context of information retrieval, which of the following

scenarios would MOST benefit from incorporating term proximity into the scoring
function?

• A query where all documents in the collection contain the query terms.

• A query consisting of a single keyword.

• A query where the order of words is irrelevant to the user's information need.

• A query targeting documents with a high term frequency for individual

keywords.

• A query where the exact phrase match is more important than individual
term occurrences.

48 | P a g e
Correct Answer: A query where the exact phrase match is more important than
individual term occurrences.

Question #128: Which of the following is the most accurate definition of tf-idf
weighting?

• A scheme that normalizes document length.

• A scheme that only considers term frequency within a document.

• A scheme that combines term frequency and inverse document frequency to

weigh terms.

• A scheme that only considers the inverse document frequency of a term.

• A scheme that only considers the length of the document.

Correct Answer: A scheme that combines term frequency and inverse document
frequency to weigh terms.

Question #129: In the context of tf-idf weighting, which of the following statements
best describes its effect on ranking documents for a query with multiple terms?

• tf-idf only considers the most frequent term in the query.

• tf-idf prioritizes documents that contain stop words.

• tf-idf gives equal weight to all terms in the query, regardless of their
frequency.

• tf-idf gives higher weight to terms that are frequent in a document but rare in
the overall collection.

• tf-idf only considers the presence or absence of query terms in a document.

Correct Answer: tf-idf gives higher weight to terms that are frequent in a document
but rare in the overall collection.

49 | P a g e
Question #130: Question: In the context of information retrieval, what is a "free text
query"?

• A query consisting of one or more words in a human language.

• A query that requires precise Boolean logic.

• A query that uses a specific query language with operators and expressions.

• A query that returns only a small subset of relevant documents.

• A query that is always interpreted literally without any stemming or

lemmatization.

Correct Answer: A query consisting of one or more words in a human language.

Question #131: What is "term frequency" (TF) in the context of information

retrieval?

• The number of documents in the collection that contain a specific term.

• The number of times a term appears in a specific document.

• The length of the term in characters.

• The number of times a term appears in the entire document collection.

• The inverse of the number of documents containing a specific term.

Correct Answer: The number of times a term appears in a specific document.

Question #132: Which of the following is the most accurate definition of "term
frequency (tf)"?

• The number of times a term appears in the entire document collection.

• The length of the term.

• The number of documents in which a term appears.

• The number of times a term appears in a specific document.

• The inverse of the number of documents in which a term appears.

50 | P a g e
Correct Answer: The number of times a term appears in a specific document.

Question #133: Which of the following best describes the trade-off between
accuracy and computational efficiency when retrieving information?

• Exact Retrieval

• Inverted Index

• Probabilistic Retrieval

• Boolean Retrieval

• Approximate Retrieval

Correct Answer: Approximate Retrieval

Question #134: Consider two search engines: Engine A returns highly relevant
results but misses some relevant documents, while Engine B returns most relevant
documents but includes some irrelevant ones. Which of the following metrics
would best differentiate the performance of these two engines, considering a user
wants to find all relevant documents?

• F1-score

• Recall@K

• Mean Reciprocal Rank (MRR)

• Precision at K (P@K)

• Mean Average Precision (MAP)

Correct Answer: Recall@K

Question #135: Which of the following is a potential drawback of query expansion?

• It requires a large amount of computational resources.

• It reduces the size of the index.

51 | P a g e
 • It can decrease the recall of the system.

• It can introduce irrelevant terms into the query, decreasing precision.

• It can only be used with the Boolean retrieval model.

Correct Answer: It can introduce irrelevant terms into the query, decreasing
precision.

Question #136: What role do static quality scores play in search engine ranking?

• They replace query-dependent scores entirely.

• They are used to eliminate stop words.

• They are combined with query-dependent scores to influence the final

ranking.

• They are only used for Boolean queries.

• They are used to normalize term frequencies.

Correct Answer: They are combined with query-dependent scores to influence the
final ranking.

Question #137: Which of the following methods, when used in index elimination,
reduces the candidate set by focusing only on documents containing many high-idf
query terms?

• Term-Document Incidence Matrix

• Inverted Index

• Boolean Retrieval

• Vector Space Model

• High-idf query terms only

Correct Answer: High-idf query terms only

52 | P a g e
Question #138: In the context of term frequency-idf (tf-idf) weighting schemes used
in information retrieval, why is cos(SaS,PaP) greater than cos(SaS,WH)?

• TF-IDF assigns higher weights to terms that appear more frequently in

documents.

• TF-IDF weights documents based on their occurrence across the corpus.

• Cosine similarity measures the angle between vectors representing text

content.

• Collection refers to the total number of documents in the dataset.

• The query is represented as a vector where each term's weight is its

frequency in the query text.

Correct Answer: TF-IDF assigns higher weights to terms that appear more
frequently in documents.

Question #139: Why is cosine similarity preferred over Euclidean distance in vector
space models for information retrieval?

• To calculate the similarity between sparse vectors

• It provides a scalar value for similarity comparison

• To measure the angle between vectors

• Because it handles documents of varying lengths effectively

• To determine document relevance based on proximity

Correct Answer: Because it handles documents of varying lengths effectively

Question #140: Why is query-term proximity important in web search?

• It ensures that all query terms are present in the document.

• It eliminates stop words.

• It improves the accuracy of stemming.

53 | P a g e
 • It reduces the computational cost of scoring.

• It indicates that the document is focused on the user's query intent.

Correct Answer: It indicates that the document is focused on the user's query
intent.

Question #141: What is a tiered index in information retrieval?

• An index that stores terms based on their length

• An index that stores terms in different languages

• An index that stores documents in different geographical locations

• An index that stores postings entries with different term frequency

thresholds

• An index that stores documents based on their file types

Correct Answer: An index that stores postings entries with different term frequency
thresholds

Question #142: Which component of a search system is responsible for generating

snippets of text that accompany each document in the results list?

• Scoring module

• Ranking module

• Document cache

• Indexer

• Query parser

Correct Answer: Document cache

Question #143: What is the main characteristic of the Bag of Words model in
information retrieval?

54 | P a g e
 • It uses stemming and lemmatization as the only preprocessing steps.

• It only considers the first 10 words of a document.

• It represents a document as an unordered set of words and their

frequencies.

• It ignores the frequency of words in a document.

• It considers the order and structure of words in a document.

Correct Answer: It represents a document as an unordered set of words and their

frequencies.

Question #144: What is the primary goal of cluster pruning?

• To reduce the size of the index

• To improve stemming efficiency

• To reduce the number of documents for which cosine scores are computed

• To eliminate duplicate documents

• To improve the accuracy of cosine similarity calculations

Correct Answer: To reduce the number of documents for which cosine scores are
computed

Question #145: What does term frequency (tf) represent?

• The number of times a term appears in a specific document.

• The inverse of the number of documents in which a term appears.

• The number of documents in which a term appears.

• The total number of terms in the collection.

• The average length of documents in the collection.

Correct Answer: The number of times a term appears in a specific document.

55 | P a g e
Question #146: How does a scoring function in information retrieval take into
account the quality of document matches?

• Jaccard coefficient

• Log-frequency weighting

• Term frequency (tf)

• Bag of words model

• Collection vs. Document frequency

Correct Answer: Collection vs. Document frequency

Question #147: Which of the following techniques is most likely to improve the
recall of an information retrieval system without significantly sacrificing precision?

• Increasing the weight of common terms in the tf-idf calculation.

• Implementing query expansion with carefully selected synonyms.

• Reducing the length of the document collection.

• Removing all stop words from the index and queries.

• Using a very aggressive stemming algorithm.

Correct Answer: Implementing query expansion with carefully selected synonyms.

Question #148: What is impact ordering in the context of information retrieval?

• Ordering query terms by decreasing order of IDF

• Ordering documents by their publication date

• Ordering documents by their static quality scores

• Ordering documents in the postings list of a term by decreasing order of tf-

idf

• Ordering documents randomly

56 | P a g e
Correct Answer: Ordering documents in the postings list of a term by decreasing
order of tf-idf

Question #149: Which of the following is NOT a typical step in processing a free text
query in an information retrieval system?

• Expanding the query with synonyms.

• Translating the query into machine code.

• Parsing the query into individual terms.

• Stemming or lemmatizing the query terms.

• Removing stop words from the query.

Correct Answer: Translating the query into machine code.

Question #150: When two documents d and d' are appended to each other, what
happens when they are length-normalized?

• The shorter document's weight is adjusted to match the longer one.

• The longer document will have a higher weight.

• Their weights will be adjusted so they have the same weight as each other.

• They will always have the highest possible cosine similarity.

• It depends on the term frequency.

Correct Answer: They will always have the highest possible cosine similarity.

Question #151: Consider two terms, 'apple' and 'the'. 'Apple' appears in 100
documents, and 'the' appears in 10,000 documents in a collection of 100,000
documents. According to inverse document frequency (idf), which term will have a
higher idf weight and why?

• The idf values cannot be compared without knowing the term frequencies
(tf).

57 | P a g e
 • Both terms will have the same idf because idf only depends on the total
number of documents.

• 'The' will have a higher idf because it appears in more documents.

• Neither term will have a significant idf because both are common words.

• 'Apple' will have a higher idf because it appears in fewer documents.

Correct Answer: 'Apple' will have a higher idf because it appears in fewer
documents.

Question #152: What is the "precision" of an information retrieval system?

• The proportion of relevant documents that are retrieved.

• The speed at which the system retrieves documents.

• The proportion of retrieved documents that are relevant.

• The amount of storage space used by the index.

• The number of documents in the collection.

Correct Answer: The proportion of retrieved documents that are relevant.

Question #153: Which data structure is typically used to efficiently select the K top
documents based on their cosine similarity scores when performing ranked
retrieval?

• Sort

• Heap

• Merge Sort

• TF-IDF

• Bag of Words

Correct Answer: Heap

58 | P a g e
Question #154: What is the main characteristic of the 'bag of words' model in
information retrieval?

• It assigns different weights to different words based on their semantic

meaning.

• It only considers the most frequent words in a document.

• It treats documents as unordered collections of words, disregarding

grammar and word order.

• It considers the order and position of words in a document.

• It uses a thesaurus to expand the query with synonyms.

Correct Answer: It treats documents as unordered collections of words,

disregarding grammar and word order.

Question #155: What does "document frequency" (DF) represent in information

retrieval?

• The number of times a specific term appears in a given document.

• The number of documents in the collection that contain a specific term.

• The frequency of the most common term in the collection.

• The average length of documents in the collection.

• The total number of terms in the document collection.

Correct Answer: The number of documents in the collection that contain a specific
term.

Question #156: Why is log-frequency weighting often used in information retrieval?

• To normalize document lengths

• To eliminate stop words

• To reduce the impact of very frequent terms

59 | P a g e
 • To improve query processing speed

• To give more weight to rare terms

Correct Answer: To reduce the impact of very frequent terms

Question #157: Which of the following is NOT a typical scale of operation for IR
systems?

• Personal information retrieval

• Enterprise search

• Quantum computing search

• Web search

• Domain-specific search

Correct Answer: Quantum computing search

Question #158: Question: What is the Soundex algorithm primarily used for?

• To translate words into different languages

• To equivalence class or expand terms with phonetic equivalents

• To generate synonyms for words

• To stem words to their root form

• To correct spelling errors in words

Correct Answer: To equivalence class or expand terms with phonetic equivalents

Question #159: Who among the following was NOT one of the main authors of the
book?

• Pavel Berkhin

• Ghulam Raza

• Thomas Zeume
60 | P a g e
 • Andrew Turner

• James Allan

Correct Answer: Pavel Berkhin

Question #160: Which of the following best describes why extended Boolean
retrieval models were used in commercial systems before the early 1990s?

• They relied solely on basic Boolean operations (AND, OR, NOT) without
additional features.

• They were primarily used after the arrival of the World Wide Web in the early
1990s.

• They were the only search method available until the invention of ranked
retrieval models.

• They provided precise control over search results using operators like
proximity (/s, /p, /k).

• They allowed users to type free text queries without any operators.

Correct Answer: They provided precise control over search results using operators
like proximity (/s, /p, /k).

Question #161: According to the text, which statement is true regarding structured
and unstructured data?

• Almost no data are truly unstructured if you consider inherent structures like
language or markup.

• Information retrieval primarily deals with structured data.

• Structured data refers exclusively to relational databases.

• All text data is considered structured.

• Unstructured data has no inherent structure at all.

61 | P a g e
Correct Answer: Almost no data are truly unstructured if you consider inherent
structures like language or markup.

Question #162: Which of the following retrieval models is most closely associated
with the concept of "relevance feedback"?

• Vector Space Model with Rocchio Algorithm

• Boolean Retrieval Model

• Inferred Model

• Language Model

• Probabilistic Retrieval Model

Correct Answer: Vector Space Model with Rocchio Algorithm

Question #163: Why might an IR system choose not to remove stop words, despite
their high frequency?

• Stop words are needed for stemming algorithms.

• Stop words are always short and easy to process.

• Stop words are essential for phrase queries.

• Stop words are crucial for language identification.

• Stop words can improve the precision of retrieval.

Correct Answer: Stop words are essential for phrase queries.

Question #164: Level 5: In a situation where all query terms have very high
document frequencies (i.e., appear in many documents), which retrieval model
might struggle the most, and why?

• Vector Space Model, because cosine similarity would be low.

62 | P a g e
 • Probabilistic Retrieval Model, because it would be difficult to estimate
probabilities.

• Boolean Retrieval, because it would return too many documents.

• All models would perform equally well.

• No model would struggle.

Correct Answer: Boolean Retrieval, because it would return too many documents.

Question #165: Question: What is the purpose of skip pointers in postings lists?

• To reduce the storage space required for postings lists

• To speed up the intersection of postings lists

• To provide a graphical representation of the document collection

• To automatically correct spelling errors in queries

• To improve the ranking of search results

Correct Answer: To speed up the intersection of postings lists

Question #166: Which of the following statements about Boolean queries is true?

• Experts recommend Boolean queries for high-recall searches.

• Using "and" in Boolean queries increases precision but may lower recall.

• Free text queries are less effective than Boolean queries.

• The extended Boolean model is better at handling spelling mistakes.

• Boolean queries are always more effective than free-text queries.

Correct Answer: Using "and" in Boolean queries increases precision but may lower
recall.

Question #167: What is the primary function of Information Retrieval (IR)?

• Locating structured data in databases

63 | P a g e
 • Organizing emails in folders

• Creating new documents

• Managing file systems

• Finding material (usually documents)

Correct Answer: Finding material (usually documents)

Question #168: What is the advantage of using a B-tree over a binary search tree for
dictionary lookup, especially when the dictionary is disk-resident?**

• B-trees don't require an ordered character set

• B-trees are always faster

• B-trees require less memory

• B-trees are easier to implement

• B-trees can collapse multiple levels, reducing disk accesses

Correct Answer: B-trees can collapse multiple levels, reducing disk accesses

Question #169: You have the wildcard query "appl*". Using a 3-gram index, which
Boolean query would you construct to find potential matching terms?

• $ap AND ppl

• apANDpl

• $ap AND apl

• apANDaplANDpl

• app AND ppl AND pl$

Correct Answer: apANDaplANDpl

Question #170: Question: What is a "positional index" used for?

• To store the citation information of documents

64 | P a g e
 • To store the geographical location of documents

• To store the sentiment of documents (positive, negative, neutral)

• To store the ranking of documents based on their relevance to a query

• To store the position of each term within a document

Correct Answer: To store the position of each term within a document

Question #171: Which retrieval model represents documents and queries as

vectors in a high-dimensional space?

• Vector Space Model

• Probabilistic Retrieval

• Inverted Index

• Language Model

• Boolean Retrieval

Correct Answer: Vector Space Model

Question #172: Level 2: What is the purpose of the "merge" operation in processing
an "AND" query using Inverted Indexes?

• To remove duplicate document IDs

• To sort the postings lists by document ID

• To calculate term frequencies

• To combine the postings lists of multiple terms

• To find the intersection of document sets containing the query terms

Correct Answer: To find the intersection of document sets containing the query
terms

65 | P a g e
Question #173: Question: In the Boolean retrieval model, how are documents
viewed?

• As a vector of term frequencies

• As a probability distribution of terms

• As a graph of interconnected words

• As a ranked list of terms

• As a set of terms

Correct Answer: As a set of terms

Question #174: What is a key advantage of using asymmetric expansion in query

processing?

• It enables

• It prevents unintended matches by allowing asymmetrical term expansions.

• It ensures that all terms are treated identically during indexing.

• It requires less processing time at query execution.

• It allows for more efficient storage by reducing the number of postings lists.

Correct Answer: It prevents unintended matches by allowing asymmetrical term

expansions.

Question #175: Which of the following is NOT a typical edit operation used to
calculate edit distance?

• Deletion

• Substitution

• Insertion

• Transposition

• Duplication

66 | P a g e
Correct Answer: Duplication

Question #176: What is the main idea behind Probabilistic Retrieval models?

• Ranking documents based on their estimated probability of relevance to the

query

• Representing documents and queries as vectors

• Retrieving documents that contain all query terms

• Using Boolean logic to retrieve documents

• Using term frequencies to rank documents

Correct Answer: Ranking documents based on their estimated probability of

relevance to the query

Question #177: How could an IR system combine the use of a positional index and
stop words? What is the potential problem, and how could it be handled? Options:

• Use positional indexes and apply stop word removal before indexing. The
potential issue is inefficiency due to multiple passes. To handle this,
optimize the process so stop words are marked during indexing.

• Integrate positional data with stop word lists for retrieval efficiency. However,
without context, this approach may be limited. Enhance it by using context-
aware algorithms.

• Leverage both methods for efficient indexing. The problem is dual

processing overhead. Use parallel processing to improve performance.

• Combine the two techniques by analyzing each word's position across

documents. The problem is that this might require complex analysis. A
solution would be to preprocess documents to mark stop words clearly.

• Merge positional and stop word data into a single index. This could cause
bloat and retrieval problems. Implement efficient storage and querying
techniques to mitigate this.

67 | P a g e
Correct Answer: Combine the two techniques by analyzing each word's position
across documents. The problem is that this might require complex analysis. A
solution would be to preprocess documents to mark stop words clearly.

Question #178: In an inverted index, what is used for vocabulary lookup to

determine if a query term exists and retrieve its postings?

• A binary search tree

• A dictionary data structure using hash tables

• An inverted list

• A term-document incidence matrix

• A wildcard query

Correct Answer: A binary search tree

Question #179: Question: What is "case-folding" in the context of text processing?

• Replacing all numbers with words

• Converting all text to lowercase

• Translating the text into a different language

• Removing all punctuation from the text

• Converting all text to uppercase

Correct Answer: Converting all text to lowercase

Question #180: Question: In determining the correct tokenization for names like
"O’Neill," which option represents the most appropriate way to tokenize this name?

• o’ neill

• neill

• o’neill

68 | P a g e
 • o neill

• oneill

Correct Answer: o’neill

Question #181: In the context of IR, what does "unstructured data" refer to?

• Data easily processed by computers

• Data organized in tables

• Data with explicit markup

• Data without clear, semantically overt structure

• Data in relational databases

Correct Answer: Data without clear, semantically overt structure

Question #182: What is the primary purpose of Retrieval Augmented Generation

(RAG) in the context of Large Language Models (LLMs)?

• To reduce the computational cost of training LLMs.

• To enable LLMs to perform sentiment analysis.

• To fine-tune LLMs for specific tasks.

• To improve the factual accuracy and reduce hallucinations by grounding the

LLM's responses in external knowledge.

• To translate text between multiple languages.

Correct Answer: To improve the factual accuracy and reduce hallucinations by

grounding the LLM's responses in external knowledge.

Question #183: Which component of the Transformer architecture allows each

element in a sequence to interact with every other element, capturing complex
contextual relationships?

69 | P a g e
 • Pooling Layer

• Recurrent Layer

• Convolutional Layer

• Embedding Layer

• Self-Attention Mechanism

Correct Answer: Self-Attention Mechanism

Question #184: In the context of prompt engineering for Large Language Models
(LLMs), what is "few-shot prompting"?

• Providing the LLM with a few examples of the desired input-output behavior
in the prompt itself.

• Training a separate model to generate prompts for the LLM.

• Prompting the LLM with very short and concise instructions.

• Providing the LLM with a large amount of training data before prompting.

• Prompting the LLM with Questions that have only a few possible answers.

Correct Answer: Providing the LLM with a few examples of the desired input-output
behavior in the prompt itself.

Question #185: What is the main advantage of using Retrieval-Augmented

Generation (RAG) with Large Language Models (LLMs)?

• It eliminates the need for fine-tuning LLMs on specific tasks.

• It reduces the computational cost of training LLMs.

• It allows LLMs to access and incorporate external knowledge sources.

• It improves the model's ability to perform mathematical calculations.

• It simplifies the process of prompt engineering.

70 | P a g e
Correct Answer: It allows LLMs to access and incorporate external knowledge
sources.

Question #186: What is the primary purpose of an inverted index in information

retrieval?

• Reduce storage space by removing zeros

• Calculate precision and recall

• Track the positions of terms within documents

• Map terms to the documents they appear in

• Store the entire term-document matrix

Correct Answer: Map terms to the documents they appear in

Question #187: Question: What is the purpose of tokenization in information

retrieval?

• To rank documents based on their relevance to a query

• To translate a document into a different language

• To remove stop words from a document

• To divide a stream of text into individual words or units

• To convert all text to lowercase

Correct Answer: To divide a stream of text into individual words or units

Question #188: What is a major challenge that LSTMs (Long Short-Term Memory
networks) are designed to address in Recurrent Neural Networks (RNNs)?

• The vanishing gradient problem, which hinders learning long-range

dependencies.

• The high computational cost of training.

71 | P a g e
 • The difficulty in parallelizing computations.

• The lack of memory capacity for storing past information.

• The inability to process sequential data.

Correct Answer: The vanishing gradient problem, which hinders learning long-
range dependencies.

Question #189: In the context of Natural Language Processing (NLP), what is a

primary limitation of the Bag of Words (BoW) approach?

• It requires significant computational resources.

• It is not applicable to text data.

• It fails to consider word frequency.

• It loses word order and grammatical structure.

• It cannot handle large vocabularies.

Correct Answer: It loses word order and grammatical structure.

Question #190: Level 3: How does stemming improve retrieval effectiveness?

• By reducing words to their root form, increasing matches

• By correcting spelling errors

• By removing common words

• By expanding the query with related terms

• By identifying synonyms

Correct Answer: By reducing words to their root form, increasing matches

Question #191: When generating multiple queries for ranked retrieval based on
user input, such as expanding "rising interest rates" into three separate phrase and

72 | P a g e
vector space queries, what is the appropriate way to aggregate the scores from
each component?

• Sum of the maximum scores from each query.

• Sum of all scores, regardless of their origin.

• Use cosine similarity alone as a combined score.

• Average of all individual scores.

• Take the highest score among all queries.

Correct Answer: Take the highest score among all queries.

Question #192: What is the key difference between document frequency (df) and
collection frequency (cf)?

• Document frequency is the number of documents in which a term appears,

while collection frequency is the total number of times a term appears in the
entire collection.

• Document frequency is calculated using logarithms, while collection

frequency is not.

• Document frequency is the number of times a term appears in a specific

document, while collection frequency is the total number of documents in
the collection.

• Document frequency and collection frequency are the same thing.

• Document frequency is used for stop words, while collection frequency is

used for rare words.

Correct Answer: Document frequency is the number of documents in which a term

appears, while collection frequency is the total number of times a term appears in
the entire collection.

Question #193: What does the term frequency (tf) represent in the context of
information retrieval?

73 | P a g e
 • The inverse of the document frequency

• The number of unique terms in the collection

• The total number of documents in the collection

• The number of times a term appears in a specific document

• The length of the document

Correct Answer: The number of times a term appears in a specific document

Question #194: What is the key difference between collection frequency and
document frequency?

• Collection frequency only considers terms in the query, while document

frequency considers terms in the documents.

• There is no difference between collection frequency and document

frequency.

• Collection frequency counts the number of documents containing a term,

while document frequency counts term occurrences in a single document.

• Collection frequency is used for ranking, while document frequency is used

for stop word removal.

• Collection frequency counts term occurrences across the entire corpus,

while document frequency counts the number of documents containing the
term.

Correct Answer: Collection frequency counts term occurrences across the entire
corpus, while document frequency counts the number of documents containing
the term.

Question #195: What is the main idea behind champion lists (or top docs) in
information retrieval?

• To cluster documents based on their similarity

74 | P a g e
 • To store all documents in a ranked order

• To eliminate documents with low IDF scores

• To precompute the top r documents for each term based on their weights

• To perform Boolean retrieval more efficiently

Correct Answer: To precompute the top r documents for each term based on their
weights

Question #196: Question: In the context of tf-idf weighting, what does the "idf"
component represent?

• The number of documents in the collection that contain a specific term.

• The length of a document normalized by the number of terms it contains.

• The inverse of the frequency of a term across the entire collection of

documents.

• The total number of terms in the collection.

• The frequency of a term within a specific document.

Correct Answer: The inverse of the frequency of a term across the entire collection
of documents.

Question #197: In the context of information retrieval, what does a "free text query"
refer to?

• A query that is free of charge.

• A query that requires the user to specify the exact location of the desired
information.

• A query that contains only stop words.

• A query that is limited to a specific number of characters.

• A query written in a natural language, without specific syntax requirements.

75 | P a g e
Correct Answer: A query written in a natural language, without specific syntax
requirements.

Question #198: In the Vector Space Model, how are documents and queries
represented?

• As sequences of characters

• As vectors in a high-dimensional space

• As nodes in a graph

• As sets of terms

• As probability distributions

Correct Answer: As vectors in a high-dimensional space

Question #199: Which of the following statements about terms and tokens in IR
systems is correct?

• All tokens are considered terms in an IR system

• Only exact matches of tokens can be considered terms

• Terms are the raw tokens before any processing occurs

• Types and terms are the same concept in IR systems

• Tokens must undergo normalization to become terms

Correct Answer: Tokens must undergo normalization to become terms

Question #200: What data structure is commonly used to efficiently find

documents containing specific terms?

• Skip List

• Term-Document Incidence Matrix

• Forward Index

76 | P a g e
 • Adjacency Matrix

• Inverted Index

Correct Answer: Inverted Index

Question #201: During a postings scan for a query q and a term t, when computing
the Jaccard coefficient, which structure is utilized to efficiently retrieve the number
of k-grams in t?

• Inverted Index

• Probabilistic Retrieval

• Term-Document Incidence Matrix

• Boolean Retrieval

• Vector Space Model

Correct Answer: Term-Document Incidence Matrix

Question #202: According to the text, what is a common heuristic for placing skip
pointers in a postings list of length L?

• Use L/2 evenly-spaced skip pointers.

• Use log(L) evenly-spaced skip pointers.

• Place a skip pointer after every term.

• Place a skip pointer at every 10th entry.

• Use √L evenly-spaced skip pointers.

Correct Answer: Use √L evenly-spaced skip pointers.

Question #203: In a B-tree used for dictionary search, what is the primary
advantage of having multiple children per internal node (compared to a binary
tree)?

77 | P a g e
 • It simplifies the rebalancing process

• It eliminates the need for a separate postings list

• It allows for faster insertion of new terms

• It makes it easier to handle wildcard queries

• It reduces the height of the tree, improving search efficiency

Correct Answer: It reduces the height of the tree, improving search efficiency

Question #204: Which of the following is TRUE about STOP WORDS in information
retrieval?

• Stop lists are used to improve search results by reducing unnecessary

postings.

• Stop words include terms like "the" and "and," which are often removed
during indexing.

• Stop words are always excluded from a vocabulary because they don't help
with document retrieval.

• Phrase searches like “President of the United States” are more affected by
stop words than keyword searches.

• Most common English words are usually considered stop words.

Correct Answer: Phrase searches like “President of the United States” are more
affected by stop words than keyword searches.

Question #205: Why is it important to apply the same normalization rules to both
indexed text and query terms?

• To make stemming more effective.

• To ensure that relevant documents are retrieved regardless of minor

variations in word form.

• To reduce the size of the inverted index.

78 | P a g e
 • To eliminate the need for stop word removal.

• To speed up the tokenization process.

Correct Answer: To ensure that relevant documents are retrieved regardless of

minor variations in word form.

Question #206: The Porter Stemmer is an example of:

• A stop word list

• A lemmatization algorithm

• A stemming algorithm

• A phrase index

• A language identification algorithm

Correct Answer: A stemming algorithm

Question #207: What is the primary data structure used for vocabulary lookup in an
inverted index?

• Array

• Binary Search Tree

• Hash Table

• Linked List

• Dictionary

Correct Answer: Dictionary

Question #208: What problem does language identification solve in Information

Retrieval?

• Translating documents

• Determining the author of a document

79 | P a g e
 • Identifying the language of a document or part of a document

• Finding documents about languages

• Detecting spam documents

Correct Answer: Identifying the language of a document or part of a document

Question #209: What is the purpose of the "Reduce" phase in the MapReduce
framework, in the context of index construction?

• To assign tasks to worker machines

• To parse the documents and extract terms

• To sort the (term, docID) pairs by term

• To distribute the document collection across multiple machines

• To merge the intermediate postings lists generated by the Map phase into
final postings lists

Correct Answer: To merge the intermediate postings lists generated by the Map
phase into final postings lists

Question #210: Which of the following best describes the relationship between
"terms" and "tokens" in Information Retrieval?

• Tokens are instances of character sequences, while terms are normalized

word types.

• Terms and tokens are synonymous.

• Tokens are used for indexing, while terms are used for query processing.

• Terms are raw, unprocessed words, while tokens are normalized words.

• Terms are always single words, while tokens can be phrases.

Correct Answer: Tokens are instances of character sequences, while terms are
normalized word types.

80 | P a g e
Question #211: Question: Which of the following is a disadvantage of the Boolean
retrieval model?

• It is difficult to implement.

• It is computationally expensive.

• It does not support phrase queries.

• It does not allow for ranked retrieval.

• It requires a large amount of storage space.

Correct Answer: It does not allow for ranked retrieval.

Question #212: Which of the following strategies is suggested for handling words
without diacritics in search queries?

• Require users to input diacritics for accurate results.

• Use machine learning models exclusively for diacritic normalization.

• Ignore the presence of accents entirely, as they are irrelevant.

• Only consider case-folding without addressing diacritics.

• Equate words regardless of diacritics to improve search accuracy.

Correct Answer: Equate words regardless of diacritics to improve search accuracy.

Question #213: What is the primary reason for breaking an inverted index into tiers
of decreasing importance?

• To ensure that documents with higher field values are retrieved before
others.

• To prevent reliance on a single point of failure in document retrieval.

• To ensure that all documents are equally represented across all tiers.

81 | P a g e
 • To improve query performance by prioritizing more important documents
first.

• To allow for easier manipulation and updating of the index.

Correct Answer: To improve query performance by prioritizing more important

documents first.

Question #214: What is the purpose of "lemmatization" in information retrieval?

• To identify the language of a document.

• To reduce words to their dictionary form (lemma).

• To expand the query with synonyms and related terms.

• To remove all punctuation from the text.

• To count the number of words in a document.

Correct Answer: To reduce words to their dictionary form (lemma).

Question #215: How does a query parser expand a user's free text query if it is of
rising interest?

• By splitting the query into subqueries based on term frequency.

• By running the query as a phrase query.

• By checking how many documents contain both phrases.

• By using vector space models to retrieve relevant documents.

• None of the above.

Correct Answer: By splitting the query into subqueries based on term frequency.

Question #216: Why is Mean Average Precision (MAP) considered a good evaluation
measure?

• It gives more weight to the first few relevant documents retrieved.

82 | P a g e
 • It only considers the top-ranked documents.

• It is easy to calculate and interpret.

• It is not affected by the number of relevant documents in the collection.

• It provides a single-figure measure of quality across recall levels.

Correct Answer: It provides a single-figure measure of quality across recall levels.

Question #217: What is the purpose of "length normalization" in the vector space
model?

• To ensure that all documents have the same number of terms.

• D) To reduce the impact of document length on similarity scores.

• To give more weight to longer documents.

• E) To remove stop words from the document collection.

• To adjust term frequencies based on document length.

Correct Answer: D) To reduce the impact of document length on similarity scores.

Question #218: What is the primary purpose of using log-frequency weighting in

information retrieval?

• To increase the impact of rare terms.

• To completely ignore frequent terms.

• To reduce the impact of very frequent terms.

• To normalize document length.

• To give more weight to frequent terms.

Correct Answer: To reduce the impact of very frequent terms.

Question #219: Why is Euclidean distance a poor choice for measuring document
similarity in the vector space model?

83 | P a g e
 • It is not sensitive to differences in term frequency.

• It only works for binary vectors.

• It is highly affected by document length.

• It cannot handle high-dimensional data.

• It is computationally expensive to calculate.

Correct Answer: It is highly affected by document length.

Question #220: When two documents d and d' (where d' is d appended to itself) are
length-normalized, what happens to their cosine similarity?

• Their cosine similarity remains the same as before normalization.

• Their cosine similarity becomes 1.

• The cosine similarity decreases because one document is longer than the
other.

• Both are now identical vectors, so their cosine similarity is 1.

• One of them becomes a unit vector but not the other.

Correct Answer: Both are now identical vectors, so their cosine similarity is 1.

Question #221: What is the main characteristic of the Bag of Words model?

• It only considers the first 100 words of a document.

• It represents a document as an ordered list of words.

• It assigns different weights to different words based on their position.

• It ignores the ordering of words in a document.

• It considers the order of words in a document.

Correct Answer: It ignores the ordering of words in a document.

84 | P a g e
Question #222: Question: Which of the following statements about the Jaccard
coefficient is FALSE?

• It measures the overlap of two sets.

• It can be used to score the similarity between a query and a document.

• It always assigns a value between 0 and 1.

• It does not require the sets being compared to be the same size.

• It considers term frequency within a document.

Correct Answer: It considers term frequency within a document.

Question #223: How can a vector space index be used to answer Boolean queries?

• By converting the Boolean query into a phrase query

• By assigning a non-zero weight to a term in the document vector if the term

occurs in the document

• Vector space indices cannot be used for Boolean queries.

• By ignoring term weights and only considering term presence

• By using a separate Boolean index

Correct Answer: By assigning a non-zero weight to a term in the document vector if

the term occurs in the document.

Question #224: Which method is typically used in information retrieval to rank

documents based on their similarity to a query?

• Jaccard coefficient

• Cosine of the angle between vectors

• Term frequency (tf)

• Euclidean distance

• Bag of Words model

85 | P a g e
Correct Answer: Cosine of the angle between vectors

Question #225: In the context of term frequency (tf), what does tft,d represent?

• The number of times that term t occurs in document d.

• The total number of terms in document d.

• The average length of documents containing term t.

• The inverse document frequency of term t.

• The number of documents in the collection that contain term t.

Correct Answer: The number of times that term t occurs in document d.

Question #226: What is a key challenge in designing an effective relevance ranking

system?

• Minimizing the storage space required for the index.

• Eliminating the need for user feedback.

• Balancing precision and recall to provide a good overall user experience.

• Ensuring that the system retrieves all documents in the collection.

• Maximizing the speed of the retrieval system, even at the expense of

accuracy.

Correct Answer: Balancing precision and recall to provide a good overall user
experience.

Question #227: The Question asks about the impact of term frequency on weight in
Information Retrieval. Question: When a term is very frequent in a collection, what
happens to its weight?

• The weight is 1 + log10(df).

• The weight decreases because it's less informative.

86 | P a g e
 • The weight increases because it becomes more common.

• The weight is always the same for all terms.

• The weight depends on the base of the logarithm used.

Correct Answer: The weight decreases because it's less informative.

Question #228: Question: Which of the following is a key property of the Jaccard
coefficient?

• It always assigns a number between 0 and 1.

• It gives higher weight to rare terms.

• It is not sensitive to document length.

• It ranges between -1 and 1.

• It considers term frequency.

Correct Answer: It always assigns a number between 0 and 1.

Question #229: Which statement about efficient cosine ranking is correct?

• Efficient cosine ranking allows us to avoid computing all N cosines and

instead select the top K documents using methods like heaps or sampling.

• Term frequency (tf) is the primary computational bottleneck in scoring.

• Document frequency determines whether a term appears in a collection,

affecting log-frequency weighting.

• To find the top K documents in a collection 'nearest' to a query, we must

compute the cosine similarity for every document.

• Jaccard coefficient is used as a replacement for cosine similarity in certain

applications.

Correct Answer: Efficient cosine ranking allows us to avoid computing all N

cosines and instead select the top K documents using methods like heaps or
sampling.

87 | P a g e
Question #230: What is the purpose of log-frequency weighting?

• To eliminate stop words.

• To reduce the impact of very frequent terms.

• To improve query processing speed.

• To normalize the length of documents.

• To increase the weight of rare terms.

Correct Answer: To reduce the impact of very frequent terms.

Question #231: Which of the following is NOT a typical component of a ranked

retrieval system?

• Returning a list of documents in no particular order.

• Indexing the document collection.

• Formulating a query in a specific query language.

• Ranking documents based on relevance to the query.

• Using term frequencies to determine document relevance.

Correct Answer: Returning a list of documents in no particular order.

Question #232: What is a limitation of using a vector space model for phrase
queries?

• The vector space model cannot assign weights to terms.

• The relative order of terms in a document is lost in the vector representation.

• The vector space model is too computationally expensive for phrase queries.

• The vector space model requires stemming, which reduces accuracy for
phrase queries.

• The vector space model cannot handle multiple terms in a query.

88 | P a g e
Correct Answer: The relative order of terms in a document is lost in the vector
representation.

Question #233: In the context of information retrieval, what is the "recall" of a

system?

• The number of documents in the collection.

• The amount of storage space used by the index.

• The speed at which the system retrieves documents.

• The proportion of retrieved documents that are relevant.

• The proportion of relevant documents that are retrieved.

Correct Answer: The proportion of relevant documents that are retrieved.

Question #234: Which of the following statements best describes the vector space
model?

• Documents are represented as sets of keywords.

• Documents and queries are represented as vectors in a high-dimensional

space.

• Documents are categorized into predefined topics.

• Documents are ranked based on the number of shared words with the query.

• Documents are represented as binary vectors.

Correct Answer: Documents and queries are represented as vectors in a high-

dimensional space.

Question #235: Which of the following best describes Ranked Retrieval?

• A model that returns an ordering of documents in a collection based on their

relevance to a free-text query, without using Boolean operators.

89 | P a g e
 • A concept where document relevance is determined by matching terms with
index words from the collection vocabulary.

• A method where systems return document rankings based on query

expressions using Boolean logic.

• A process that calculates the likelihood of documents being relevant based

on their occurrence in the text.

• A technique involving term frequency analysis to determine document

weights for scoring.

Correct Answer: A model that returns an ordering of documents in a collection

based on their relevance to a free-text query, without using Boolean operators.

Question #236: Which of the following is the first step in basic crawler operation?

• Beginning with known seed URLs.

• Securing the network.

• Designing the user interface.

• Analyzing website traffic.

• Creating a database schema.

Correct Answer: Beginning with known seed URLs.

Question #237: When performing a basic search using Boolean operators, field
codes, truncation (*), and wildcard (?) symbols, how are these symbols used to
handle word boundaries in a phrase search?

• Truncation (*) is used to limit the search to exact phrases by enclosing them
in quotes.

• The asterisk (*) symbol is used to indicate the exclusion of certain terms
during a search.

90 | P a g e
 • Phrase searching requires using both truncation (*) and wildcard (?) symbols
simultaneously.

• Wildcard (?) can be used to replace any character within a word when
searching for partial matches.

• Truncation (*) and wildcard (?) symbols are used interchangeably without
considering the need for exact phrases.

Correct Answer: Wildcard (?) can be used to replace any character within a word
when searching for partial matches.

Question #238: When a search term is entered without a field label in PubMed,
what happens if automatic term mapping fails?

• The search term is added to the query box.

• The search is canceled and redisplayed.

• The entire phrase is treated as a single search term.

• The search returns no results.

• The term is parsed into individual words for separate mapping.

Correct Answer: The term is parsed into individual words for separate mapping.

Question #239: What does Google's crawler do when encountering cached content
from potentially malicious pages?

• Alerts users about the possibility of malicious content.

• Does not distinguish between safe and unsafe cached content.

• Uses cached content to determine relevance.

• Deletes cached content if it's deemed unsafe.

• Automatically ignores cached content.

Correct Answer: Automatically ignores cached content.

91 | P a g e
Question #240: You are comparing two search engines, A and B, using MAP. Engine
A has a MAP of 0.8, while Engine B has a MAP of 0.6. However, for a specific query,
Engine B returns a perfectly relevant result at rank 1, while Engine A returns its first
relevant result at rank 3. Which engine is better for this specific query from a user
perspective, and why?

• Engine A, because it has a higher overall MAP score.

• Engine A, because MAP considers both precision and recall.

• Engine B, because it provides a perfectly relevant result at the top.

• It cannot be determined without knowing the total number of relevant

documents.

• The two engines perform equally well for this query.

Correct Answer: Engine B, because it provides a perfectly relevant result at the top.

Question #241: Which of the following features is included as part of PubMed's

explicit politeness?

• Query box entry for search terms

• Inverted index preview of search results

• History of previous searches

• Limits by date, gender, age, language, or field label

• Clipboard manipulation

Correct Answer: Query box entry for search terms

Question #242: How does EBSCO Host manage its URL frontier?

• It employs both adjacency searching and exact phrase matching for URL
storage.

• It stores cached pages for faster access during repeated searches.

92 | P a g e
 • It uses expanders to refine search results without requiring quotation marks.

• It dynamically generates URLs based on search terms and contexts.

• It segments URLs into hierarchical levels using a scheme similar to the

Mercator system.

Correct Answer: It segments URLs into hierarchical levels using a scheme similar
to the Mercator system.

Question #243: Which of the following scenarios would MRR be the MOST
appropriate evaluation metric?

• Evaluating a search engine designed to retrieve all relevant research papers

for a given topic.

• Evaluating a search engine designed to provide a diverse set of results for a

broad query.

• Evaluating a search engine designed to rank products based on customer

reviews.

• Evaluating a search engine designed to find the homepage of a specific

company, given its name.

• Evaluating a search engine designed to provide a comprehensive overview of

a historical event.

Correct Answer: Evaluating a search engine designed to find the homepage of a

specific company, given its name.

Question #244: What is the primary function of a web crawler?

• To create social media posts.

• To systematically discover, retrieve, and index web content.

• To manage network infrastructure.

• To display web pages to users.

93 | P a g e
 • To design website layouts.

Correct Answer: To systematically discover, retrieve, and index web content.

Question #245: Does EBSCO Host save search results for users by default, or does
it require a user to sign in to retain saved searches?

• Searches are stored permanently once a user signs up.

• It uses quotation marks for exact phrase searches without needing limiters.

• It saves all searches in the user's session unless they log out.

• It does not save searches automatically and requires a login for any saved
results.

• It allows users to export or share search results via email.

Correct Answer: It saves all searches in the user's session unless they log out.

Question #246: Which three refinement options are available through the tab on
the results screen?

• Date ranges, subject categories, and appearance in forums

• Date ranges, peer-reviewed journals, and full-text availability

• Date ranges, appearance in blogs, and availability in books

• Date ranges, citation indexes, and availability in PDFs

• Date ranges, appearance in news articles, and availability in conference

papers

Correct Answer: Date ranges, peer-reviewed journals, and full-text availability

Question #247: What does it mean for a web crawler to be "extensible"?

• It can only crawl a fixed number of web pages.

• It can adapt to new data formats and protocols.

94 | P a g e
 • It can only crawl websites with specific domain extensions.

• It can automatically generate website content.

• It can operate without any human intervention.

Correct Answer: It can adapt to new data formats and protocols.

Question #248: When performing an advanced search on EBSCO's platform, which

of the following options allows users to refine their search by limiting results based
on specific criteria such as date ranges or full-text availability?

• Advanced search with field labels and Boolean operators

• Truncation and wildcards without Boolean operators

• Field codes only

• A dedicated refinement tab that includes date ranges and peer-reviewed

journals

• Automatic synonym inclusion without user intervention

Correct Answer: A dedicated refinement tab that includes date ranges and peer-
reviewed journals

Question #249: Consider two search engines. Engine X has a higher MAP score
than Engine Y on a given test collection. What can we generally infer from this?

• Engine X is guaranteed to return more relevant documents than Engine Y for

every query.

• Engine X is likely to provide better overall ranking performance than Engine Y

across a range of queries.

• Engine X is faster than Engine Y.

• Engine X is always better than Engine Y for all possible queries.

• Engine X is more user-friendly than Engine Y.

95 | P a g e
Correct Answer: Engine X is likely to provide better overall ranking performance
than Engine Y across a range of queries.

Question #250: When a search term is entered into PubMed without specifying a
field label, how does the system handle the search?

• Looks only for MeSH (Medical Subject Headings) terms.

• Does not consider any field labels or controlled vocabularies.

• Searches for matches in the MeSH table, then explodes more specific
subheadings if available.

• Searches first in the Journals table, then in the Phrase List.

• Searches all fields and uses AND operations after failing to translate the
term.

Correct Answer: Searches for matches in the MeSH table, then explodes more
specific subheadings if available.

Question #251: Which of the following metrics is used to evaluate unranked

retrieval?

• Click-Through Rate (CTR).

• Precision@K (P@K).

• Mean Average Precision (MAP).

• Mean Reciprocal Rank (MRR).

• Precision.

Correct Answer: Precision.

Question #252: What is a key characteristic of relevance in the context of search

engine evaluation?

96 | P a g e
 • Relevance is solely determined by the number of keywords in a document
that match the query.

• Relevance is only important for academic search engines, not for

commercial ones.

• Relevance is subjective and can vary from person to person.

• Relevance is static and does not change over time.

• Relevance is purely objective and can be determined by algorithms alone.

Correct Answer: Relevance is subjective and can vary from person to person.

Question #253: what PubMed does after a search term is entered without a field
labe?

• Explodes the term using MeSH hierarchy.

• Searches only in the Author Index.

• Uses an automatic mapping feature. From the context, PubMed uses an

automatic term mapping feature that includes exploding the term via MeSH
hierarchy if applicable. Therefore, the Correct Answer is:

• Displays results in batches of 20.

• Translates the term into other languages.

Correct Answer: Explodes the term using MeSH hierarchy.

Question #254: A crawler is designed to fetch pages of higher quality first. Which of
the following metrics would be most effective in estimating page quality before
fetching?

• The number of images on the page.

• The number of inbound links to the page from known reputable sites.

• The number of outbound links on the page.

• The length of the page's HTML code.

97 | P a g e
 • The page's server response time.

Correct Answer: The number of inbound links to the page from known reputable
sites.

Question #255: A web crawler is tasked with continuously updating its index.
Which of the following strategies would best balance freshness and resource
usage?

• Never recrawling pages to conserve resources.

• Only recrawling pages that have been manually flagged for updates.

• Prioritizing recrawling based on the estimated change frequency of each

page.

• Recrawling all pages every 24 hours.

• Randomly recrawling a small subset of pages each day.

Correct Answer: Prioritizing recrawling based on the estimated change frequency

of each page.

Question #256: In information retrieval, when formulating a query that involves

multiple terms, what is an effective way to reduce the number of documents
considered during retrieval?

• Include only documents that contain many query terms.

• Include all documents containing at least one query term.

• Only include documents containing all query terms.

• Include only high-idf query terms.

• Exclude documents based on collection frequency rather than document

frequency.

Correct Answer: Include only documents that contain many query terms.

98 | P a g e
Question #257: Consider a query "natural language processing". Document A has a
high term frequency for "language" but a low term frequency for "natural" and
"processing". Document B has moderate term frequencies for all three terms.
Assuming idf values are similar for all three terms, which document is likely to be
ranked higher by tf-idf and why?

• Document B, because it has moderate term frequencies for all the query
terms.

• Document A, because tf-idf prioritizes documents with the highest overall

term frequency, regardless of term distribution.

• Document B, because tf-idf always favors documents with shorter lengths.

• Document A, because it has a high term frequency for at least one of the
terms.

• Neither, because the idf values are similar, so the ranking will be arbitrary.

Correct Answer: Document B, because it has moderate term frequencies for all the
query terms.

Question #258: Consider two documents, D1 and D2. D1 contains the term
"information" 5 times and D2 contains it 10 times. The document frequency of
"information" in the collection is 100 (meaning 100 documents contain the term).
Assuming a collection size of 10,000 documents, which document would receive a
higher tf-idf score for the term "information" before length normalization?

• Both would have the same score because tf-idf balances term frequency
and document frequency.

• D2, because it has a higher term frequency.

• Neither, because "information" is likely a stop word.

• It is impossible to determine without knowing the average document length.

• D1, because it is shorter.

Correct Answer: D2, because it has a higher term frequency.

99 | P a g e
Question #259: Which retrieval model assigns higher weights to terms based on
their rarity and document frequency?

• Inverted Index

• Vector Space Model

• Term-Document Incidence Matrix

• Boolean Retrieval

• Probabilistic Retrieval

Correct Answer: Probabilistic Retrieval

Question #260: Which of the following is a potential issue when using stemming in
information retrieval?

• It can increase the size of the index.

• It can lead to over-stemming.

• It requires a large amount of computational resources.

• It decreases the recall of the system.

• It can only be used with the Boolean retrieval model.

Correct Answer: It can lead to over-stemming.

Question #261: What is the primary goal of ranked retrieval models in Information
Retrieval?

• To return a random selection of documents from the collection.

• To return an ordered list of documents based on their relevance to the query.

• To return only the documents that exactly match the query.

• To return all documents that contain the query terms.

• To exclude documents that contain stop words.

100 | P a g e
Correct Answer: To return an ordered list of documents based on their relevance to
the query.

Question #262: Which of the following is a potential problem with using only term
frequency (TF) to rank documents?

• It favors longer documents over shorter ones.

• It gives too much weight to rare terms.

• It is computationally expensive to calculate.

• It cannot be used with stemming.

• It gives too little weight to common terms.

Correct Answer: It favors longer documents over shorter ones.

Question #263: Which of the following statements correctly describes the term
weighting process used in information retrieval?

• Inverted index structures are used to store term positions, not term weights.

• Term weights increase with both the frequency of the term in a document
and its rarity in the collection.

• The weight for a term in a document is the product of its frequency and
inverse document frequency.

• Term weights are calculated using a simple addition model: w_t,d = 1 +

log10(tf_t,d).

• Term weights are determined by the number of occurrences within a

document, without considering document frequency.

Correct Answer: The weight for a term in a document is the product of its
frequency and inverse document frequency.

Question #264: How does idf affect the ranking of documents for queries?

101 | P a g e
 • It only affects the ranking of documents for one-term queries.

• It always boosts the score of documents containing rare terms.

• It always lowers the score of documents containing frequent terms.

• It has no effect on ranking.

• It affects the ranking of documents for queries with at least two terms.

Correct Answer: It affects the ranking of documents for queries with at least two
terms.

Question #265: In the context of search engine design, what is evidence

accumulation?

• The process of normalizing term frequencies

• The process of combining evidence of a document's relevance from multiple

sources

• The process of stemming query terms

• The process of indexing documents based on their publication date

• The process of eliminating irrelevant documents from the search results

Correct Answer: The process of combining evidence of a document's relevance

from multiple sources

Question #266: Which of the following best describes the Jaccard coefficient?

• A measure of the similarity between two sets.

• A measure of term frequency in a document.

• A measure of the dissimilarity between two documents.

• A measure of the number of shared terms between two documents.

• A measure of the angle between two document vectors.

Correct Answer: A measure of the similarity between two sets.

102 | P a g e
Question #267: Which of the following is a component of the Jaccard coefficient
used for measuring the similarity between two sets?

• Number of common elements

• Difference between the sets

• Intersection of the sets

• Union of the sets

• Size of the individual sets

Correct Answer: Intersection of the sets

Question #268: Question: Which of the following is a primary characteristic of the

"bag of words" model in information retrieval?

• It represents documents as ordered sequences of words.

• It ignores the order of words, treating documents as unordered collections of

terms.

• It requires a positional index to function correctly.

• It assigns different weights to different word positions in a document.

• It considers the order of words in a document.

Correct Answer: It ignores the order of words, treating documents as unordered

collections of terms.

Question #269: In the bag of words model, what aspect of a document is

disregarded?

• Term order

• Rare words

• Term frequency

103 | P a g e
 • Document length

• Stop words

Correct Answer: Term order

Question #270: You are designing a search engine for a very large, dynamic
collection of documents (like the web) where the vocabulary is constantly growing.
You need to balance search speed, storage space, and the ability to handle both
wildcard queries and spelling correction. Which combination of data structures and
techniques would offer the BEST overall compromise?

• B-tree for the dictionary, k-gram index for wildcards, edit distance for
spelling correction, and a mechanism for periodically rebuilding the k-gram
index.

• Hash table for the dictionary, permuterm index for wildcards, and a
combination of edit distance and phonetic matching for spelling correction.

• Hash table for the dictionary, k-gram index for wildcards, and Soundex for
spelling correction.

• B-tree for the dictionary, permuterm index for wildcards, and edit distance
for spelling correction.

• Trie for the dictionary, a combination of k-gram and permuterm indexes for
wildcards, and a probabilistic spelling correction model.

Correct Answer: B-tree for the dictionary, k-gram index for wildcards, edit distance
for spelling correction, and a mechanism for periodically rebuilding the k-gram
index.

Question #271: Which of the following is NOT a characteristic of ranked retrieval

models?

• Documents are ordered based on their relevance to the query.

• Ranked retrieval is often associated with free text queries.

104 | P a g e
 • Free text queries are typically used.

• Users are required to use a specific query language with operators and
expressions.

• The system returns the top documents in the collection for a query.

Correct Answer: Users are required to use a specific query language with operators
and expressions.

Question #272: Which of the following is a characteristic of a "ranked" retrieval

model compared to a "Boolean" retrieval model?

• Boolean retrieval allows for partial matches.

• Boolean retrieval is more efficient for large document collections.

• Ranked retrieval only works with structured data.

• Ranked retrieval assigns a relevance score to each document.

• Ranked retrieval returns only exact matches to the query.

Correct Answer: Ranked retrieval assigns a relevance score to each document.

Question #273: Question: What is the purpose of a biword index?

• To provide a visual representation of word relationships

• To efficiently process phrase queries consisting of two words

• To index documents written in two languages

• To store biographical information about authors

• To index documents containing only two words

Correct Answer: To efficiently process phrase queries consisting of two words

Question #274: Question: Which of the following is a consideration when choosing

a document unit for indexing?

105 | P a g e
 • The number of images in the document

• The author of the document

• The granularity of indexing (e.g., chapter, paragraph, sentence)

• The font size used in the document

• The color scheme of the document

Correct Answer: The granularity of indexing (e.g., chapter, paragraph, sentence)

Question #275: In the context of information retrieval, what does "relevance

ranking" refer to?

• Presenting documents in the order they were added to the collection.

• Ordering documents based on their publication date.

• Assigning random scores to documents.

• Ordering documents alphabetically by title.

• Ordering documents based on their estimated relevance to a user's query.

Correct Answer: Ordering documents based on their estimated relevance to a

user's query.

Question #276: Which statement best describes Calvin Mooers' view on the use of
Boolean algebra in information retrieval system design?

• Calvin Mooers strongly disagreed with the use of Boolean algebra for
retrieval systems.

• Calvin Mooers focused on improving indexing structures rather than

formalisms like Boolean algebra.

• Calvin Mooers advocated for the use of other formalisms instead of Boolean
algebra.

• Calvin Mooers believed Boolean algebra was the most appropriate

formalism for retrieval system design.

106 | P a g e
 • Calvin Mooers supported the use of Boolean algebra due to its efficiency in
retrieval systems.

Correct Answer: Calvin Mooers strongly disagreed with the use of Boolean algebra
for retrieval systems.

Question #277: How does Japanese typically handle inflectional endings and
function words?

• Hiragana

• Arabic numerals

• Kanji

• Katakana

• Latin letters

Correct Answer: Hiragana

Question #278: Question 2:** In a standard inverted index, what does each entry in
the postings list represent?

• A rotated version of a term

• The frequency of a term in the document

• A document containing a specific term

• A term in the vocabulary

• The phonetic representation of the term

Correct Answer: A document containing a specific term

Question #279: When trying to efficiently retrieve all vocabulary terms that begin
with a specific prefix (e.g., "automat"), why is a binary search tree considered
superior to hashing for this task?

107 | P a g e
 • Hashing allows quick lookups but cannot easily handle prefix-based
searches.

• Hashing is better suited for dynamic updates where keys can be added or
deleted frequently.

• Hashing uses an inverted index, which inherently supports prefix searches.

• Search trees permit efficient retrieval of terms based on partial information

without collisions.

• Binary search trees require more memory compared to hashing structures.

Correct Answer: Hashing allows quick lookups but cannot easily handle prefix-
based searches.

Question #280: Question: In the context of information retrieval, what is a "term"?

• A normalized token that is included in the IR system’s dictionary

• Any word in a document

• A document that is relevant to a query

• A statistical measure of a word's importance

• A graphical representation of word relationships

Correct Answer: A normalized token that is included in the IR system’s dictionary

Question #281: When using a k-gram index for spelling correction, why is a post-
filtering step often necessary after retrieving potential matches using a Boolean
query on the k-grams?

• To ensure that the matched terms have the correct phonetic representation

• To handle wildcard characters within the query term

• To remove terms that do not contain all the k-grams of the query

• To calculate the edit distance between the query and the potential matches

108 | P a g e
 • To remove terms which, although contain the required k-grams, do not
match the original query

Correct Answer: To remove terms which, although contain the required k-grams,
do not match the original query

Question #282: An IR system is experiencing performance bottlenecks during

query processing due to very long postings lists for common terms. Which of the
following strategies would be MOST effective in addressing this issue, assuming a
relatively static index?

• Increasing the aggressiveness of stemming.

• Switching from an inverted index to a term-document matrix.

• Removing more stop words from the index.

• Using a larger vocabulary size.

• Implementing skip pointers in the postings lists.

Correct Answer: Implementing skip pointers in the postings lists.

Question #283: What is the main disadvantage of using a permuterm index for
wildcard queries?

• It results in a significant increase in dictionary size

• It is slower than using a B-tree

• It cannot handle leading wildcard queries

• It requires a phonetic hashing algorithm

• It cannot handle queries with multiple wildcards

Correct Answer: It results in a significant increase in dictionary size

Question #284: Which of the following pairs of words would you argue should NOT
be conflated by the Porter stemmer, and why?

109 | P a g e
 • absorbency/absorbent

• abandon/abandonment

• marketing/markets

• volume/volumes

• university/universe

Correct Answer: volume/volumes

Question #285: Why is performance/efficiency important for a web crawler?

• To ensure that all crawled pages are visually appealing.

• To prevent users from accessing crawled data.

• To reduce the number of web pages crawled.

• To minimize the amount of data stored.

• To permit full use of available processing and network resources.

Correct Answer: To permit full use of available processing and network resources.

Question #286: System A returns the following ranked list for a query (R = Relevant,
N = Non-relevant): R, N, N, R, N, R. System B returns: N, R, N, R, N, R. Which system
has a higher Precision@3?

• Precision@3 is not a valid metric in this scenario.

• System A.

• Both systems have the same Precision@3.

• It cannot be determined without knowing the total number of relevant

documents.

• System B.

Correct Answer: System A.

110 | P a g e
Question #287: What is the difference between collection frequency (cf) and
document frequency (df)?

• cf is the total number of times a term appears in the collection, while df is

the number of documents in which the term appears.

• cf is the number of query terms, while df is the number of documents

retrieved.

• cf is the number of times a term appears in a document, while df is the

number of documents in which the term appears.

• cf is the average term frequency, while df is the median term frequency.

• cf is the number of characters in the collection, while df is the number of

documents in the collection.

Correct Answer: cf is the total number of times a term appears in the collection,
while df is the number of documents in which the term appears.

Question #288: In the bag of words model, what is ignored?

• The relevance of a term to the query.

• The total number of terms in the document.

• The presence or absence of a term in a document.

• The number of times a term appears in a document.

• The order of terms in a document.

Correct Answer: The order of terms in a document.

Question #289: In a vector space model, what does the "cosine similarity"
measure?

• The number of words two documents have in common.

• The angle between the vectors representing two documents.

• The number of documents in the collection that contain a specific term.

111 | P a g e
 • The length of the vectors representing two documents.

• The Euclidean distance between the vectors representing two documents.

Correct Answer: The angle between the vectors representing two documents.

Question #290: In Information Retrieval, what is a primary purpose of maintaining

high and low lists when processing a query?

• To cluster documents using random sampling techniques

• To partition the document collection into two tiers for efficient retrieval

• To early terminate the search process based on specific thresholds

• To prioritize higher quality documents during retrieval

• To manage impact-ordered postings for score computation

Correct Answer: To partition the document collection into two tiers for efficient
retrieval

Question #291: Question: Why is raw term frequency (tf) not ideal for directly
computing query-document match scores?

• It doesn't consider the importance of rare terms.

• It doesn't account for the length of the document.

• Relevance does not increase proportionally with term frequency.

• It is difficult to calculate for large documents.

• It requires normalization to be used effectively.

Correct Answer: Relevance does not increase proportionally with term frequency.

Question #292: In the context of information retrieval, what does "idf" stand for?

• Iterative Document Filter.

• Integrated Data Feed.

112 | P a g e
 • Item Data Format.

• Indexed Data Filter.

• Inverse Document Frequency.

Correct Answer: Inverse Document Frequency.

Question #293: What is the main purpose of using "tf-idf" weighting in information
retrieval?

• To favor documents with shorter lengths.

• To ignore the frequency of terms in documents.

• To give all terms equal importance.

• To only consider the first occurrence of each term in a document.

• To give higher weights to terms that are frequent in a specific document but
rare in the overall document collection.

Correct Answer: To give higher weights to terms that are frequent in a specific
document but rare in the overall document collection.

Question #294: Which of the following retrieval models uses a method where each
term is associated with two postings lists, referred to as "high" and "low"?

• Cluster Pruning

• Probabilistic Retrieval

• Boolean Retrieval

• Vector Space Model

• Term-Document Incidence Matrix

Correct Answer: Vector Space Model

113 | P a g e
Question #295: What is the purpose of using "tf-idf normalization" in information
retrieval?

• To adjust term frequencies based on document length and term rarity.

• To remove stop words from the document collection.

• To ensure that all documents have the same length.

• To translate the query into a different language.

• To give more weight to common terms.

Correct Answer: To adjust term frequencies based on document length and term
rarity.

Question #296: What is the primary reason for using "inverse document frequency"
(IDF) in information retrieval?

• To decrease the weight of rare terms.

• To normalize the length of documents.

• To increase the weight of terms that appear in many documents.

• To decrease the weight of terms that appear in many documents.

• To increase the weight of common terms.

Correct Answer: To decrease the weight of terms that appear in many documents.

Question #297: In the context of Information Retrieval, Term Frequency (tf) plays a
crucial role in which model?

• No, it isn't part of tf-idf weighting.

• Yes, it is.

• It is calculated using log10(1 + tf,d).

• It is the same as inverse document frequency (idf).

• It depends on the number of documents, not the collection.

114 | P a g e
Correct Answer: Yes, it is.

Question #298: What does Mean Reciprocal Rank (MRR) primarily measure?

• The cost of retrieving relevant documents.

• The average recall across all queries.

• The average precision across all queries.

• How quickly the system finds the first relevant document.

• The number of relevant documents retrieved per query.

Correct Answer: How quickly the system finds the first relevant document.

Question #299: A web crawler is designed to respect website owners' preferences

while maximizing crawl coverage. Which of the following approaches would best
balance these competing goals?

• Only crawling websites that explicitly grant permission via a custom API.

• Using a combination of [Link] adherence, crawl delay settings, and user-

agent identification to adapt to each website's specific preferences.

• Ignoring [Link] directives to ensure complete coverage, but throttling

crawl rates to avoid overloading servers.

• Crawling all pages at a fixed rate, regardless of website preferences.

• Always obeying [Link] directives and never crawling any disallowed

pages.

Correct Answer: Using a combination of [Link] adherence, crawl delay settings,

and user-agent identification to adapt to each website's specific preferences.

Question #300: Which of the following best describes the Mercator scheme in the
context of a URL frontier?

115 | P a g e
 • A system for partitioning URLs across multiple crawlers to improve
efficiency.

• A method for automatically translating web pages into different languages.

• A method for visualizing crawled data on a map.

• A technique for detecting and avoiding spider traps.

• A strategy for prioritizing URLs based on geographical location.

Correct Answer: A system for partitioning URLs across multiple crawlers to

improve efficiency.

Question #301: How is Average Precision (AP) calculated for a single query when
evaluating ranked retrieval?

• By summing the precision values at each rank where a relevant document is

retrieved, and then dividing by the total number of relevant documents.

• By summing the precision values at each rank where a relevant document is

retrieved, and then dividing by the total number of retrieved documents.

• By calculating the reciprocal rank of the first relevant document.

• By calculating the precision at K, where K is the number of relevant

documents.

• By averaging the precision values at each rank.

Correct Answer: By summing the precision values at each rank where a relevant
document is retrieved, and then dividing by the total number of relevant
documents.

Question #302: What is Precision@K (P@K)?

• The number of relevant documents retrieved divided by K.

• Precision calculated considering only the top K documents retrieved.

• Precision calculated only on the first K relevant documents retrieved.

• The average precision of the top K documents across all queries.

116 | P a g e
 • The recall value when only K documents are retrieved.

Correct Answer: Precision calculated considering only the top K documents

retrieved.

Question #303: You are evaluating a search engine designed for Question
answering. Which metric would be most appropriate to prioritize?

• Mean Reciprocal Rank (MRR).

• Mean Average Precision (MAP).

• Precision@10 (P@10).

• F1-measure.

• Recall.

Correct Answer: Mean Reciprocal Rank (MRR).

Question #304: Which of the following is a key advantage of the Transformer

architecture over Recurrent Neural Networks (RNNs) for sequence processing?

• Transformers can capture long-range dependencies more effectively and

can be parallelized, while RNNs struggle with both.

• Transformers are more susceptible to the vanishing gradient problem than

RNNs.

• Transformers are generally more difficult to train and require more data than
RNNs.

• Transformers can only process shorter sequences than RNNs.

• Transformers require sequential processing of input data, while RNNs can

process data in parallel.

Correct Answer: Transformers can capture long-range dependencies more

effectively and can be parallelized, while RNNs struggle with both.

117 | P a g e
Question #305: Which of the following best describes the "scaling hypothesis" in
the context of Large Language Models (LLMs)?

• The hypothesis that increasing the amount of training data always leads to
better model performance, regardless of model size.

• The hypothesis that scaling the number of layers in the model is more
important than scaling the number of parameters per layer.

• The hypothesis that increasing model size dramatically improves

performance, especially when combined with more training data.

• The hypothesis that scaling the learning rate during training can lead to
faster convergence and better results.

• The hypothesis that scaling down the model size can improve efficiency
without sacrificing performance.

Correct Answer: The hypothesis that increasing model size dramatically improves
performance, especially when combined with more training data.

Question #306: Which of the following is a key characteristic of Supervised

Learning?

• Learning through interaction with an environment

• Discovering hidden patterns in data

• Action-reward feedback mechanism

• Learning from labeled training data

• Learning from unlabeled data

Correct Answer: Learning from labeled training data

Question #307: What is the vanishing gradient problem in RNNs, and why does it
occur?

• Gradients become exponentially larger, causing unstable training.

118 | P a g e
 • Gradients become extremely small, preventing learning long-range
dependencies.

• The model overfits to the training data, reducing generalization.

• The learning rate is too high, causing oscillations in the loss function.

• The input data is not properly normalized, leading to numerical instability.

Correct Answer: Gradients become extremely small, preventing learning long-

range dependencies.

Question #308: In EBSCO Host's basic search screen, what does typing a term into
the Find box do?

• It performs an adjacency search for multiple words as they appear in the

text.

• It searches for exact phrases without using quotation marks.

• It requires quotation marks to perform an exact phrase search.

• It saves the search for future use.

• It checks for misspellings and offers corrections.

Correct Answer: It searches for exact phrases without using quotation marks.

Question #309: What is the purpose of "Crawler Seed Pages"?

• To define the visual layout of a website.

• To store user data collected by the crawler.

• To encrypt web traffic.

• To act as the initial URLs for a web crawler to begin exploration.

• To prevent crawlers from accessing certain websites.

Correct Answer: To act as the initial URLs for a web crawler to begin exploration.

119 | P a g e
Question #310: Which of the following is true regarding collection frequency and
document frequency?

• Collection frequency and document frequency are always equal.

• A term's collection frequency is always less than or equal to its document

frequency.

• Document frequency only considers unique terms, while collection

frequency considers all terms.

• Collection frequency only considers unique terms, while document

frequency considers all terms.

• A term's document frequency is always less than or equal to its collection

frequency.

Correct Answer: A term's document frequency is always less than or equal to its
collection frequency.

Question #311: In the context of information retrieval, what is "query expansion"?

• Adding terms to the user's query to broaden the search.

• Removing stop words from the query.

• Automatically correcting spelling errors in the query.

• Translating the query into a different language.

• Reducing the length of the user's query.

Correct Answer: Adding terms to the user's query to broaden the search.

Question #312: Which of the following best describes how queries are represented
in a Vector Space Model?

• Queries are ranked based on their similarity to documents.

• Queries are treated as binary indicators of presence or absence of terms.

• Queries use only TF weights without considering IDF.

120 | P a g e
 • Queries are matched directly to documents using an inverted index.

• Queries are converted into numerical vectors where each term is mapped to
a unique dimension.

Correct Answer: Queries are converted into numerical vectors where each term is
mapped to a unique dimension.

Question #313: Imagine you are designing an information retrieval system for a
highly specialized domain, such as patent law, where the cost of missing a relevant
document is extremely high. How would you prioritize the various evaluation
metrics (precision, recall, F-measure, MAP, R-precision, NDCG) and system
characteristics (indexing speed, search latency, index size) to optimize the system
for this specific application? Explain your reasoning for each prioritization.

• Prioritize index size, followed by indexing speed and search latency.

Precision and recall are secondary. MAP and NDCG are most important for
optimizing ranking.

• Prioritize recall above all else, followed by precision. Indexing speed and
search latency are secondary but important. Index size is least important. F-
measure and R-precision are useful for balancing precision and recall, while
MAP and NDCG are less relevant due to the high cost of missing relevant
documents.

• Prioritize indexing speed and search latency, followed by a balance between

precision and recall. Index size is least important. F-measure is the primary
metric for evaluating the balance.

• Prioritize a balance between precision and recall, using the F-measure as

the primary metric. Search latency and indexing speed are secondary. Index
size is least important. MAP and R-precision are useful for optimizing the
balance.

• Prioritize precision above all else, followed by recall. Search latency is

secondary but important. Index size and indexing speed are least important.
MAP and NDCG are most important for optimizing ranking.

121 | P a g e
Correct Answer: Prioritize recall above all else, followed by precision. Indexing
speed and search latency are secondary but important. Index size is least
important. F-measure and R-precision are useful for balancing precision and recall,
while MAP and NDCG are less relevant due to the high cost of missing relevant
documents.

Question #314: Which of the following is a key limitation of the Bag of Words (BoW)
model?

• It is computationally expensive to implement.

• It is unable to capture semantic relationships between words.

• It is only applicable to short documents.

• It requires large amounts of training data.

• It cannot be used for text classification tasks.

Correct Answer: It is unable to capture semantic relationships between words.

Question #315: When a page is identified as malicious or spam, what action does
Google's crawler take?

• The crawler uses cached content when the original page is unavailable but
still indexes it.

• The crawler dynamically generates spider traps to ensnare the site.

• The crawler applies implicit politeness rules before crawling.

• The crawler only accesses pages on the URL frontier using the Mercator
scheme.

• The crawler skips indexing the page and doesn't cache any links from it.

Correct Answer: The crawler skips indexing the page and doesn't cache any links
from it.

122 | P a g e
Question #316: Which unsupervised learning algorithm partitions data into K
clusters by iteratively assigning points to the nearest centroid and updating
centroids based on cluster means?

• Principal Component Analysis (PCA)

• Autoencoders

• K-Means Clustering

• Linear Regression

• Hierarchical Clustering

Correct Answer: K-Means Clustering

Question #317: What is "implicit politeness" in the context of web crawling?

• Always crawling a website at the same time each day.

• Respecting server resources by avoiding hitting any site too often, even
without explicit instructions.

• Ignoring [Link] to ensure complete coverage.

• Only crawling websites that use a specific type of encryption.

• Crawling only government websites.

Correct Answer: Respecting server resources by avoiding hitting any site too often,
even without explicit instructions.

Question #318: What is the purpose of positional encoding in the Transformer

architecture?

• To enable the model to process the input sequence in a sequential manner.

• To normalize the input data to improve training stability.

• To provide the model with information about the meaning of words in the
input sequence.

123 | P a g e
 • To inject information about the position of tokens in the input sequence, as
self-attention is order-agnostic.

• To reduce the dimensionality of the input embeddings.

Correct Answer: To inject information about the position of tokens in the input
sequence, as self-attention is order-agnostic.

Question #319: Which of the following best describes the key difference between
BERT and GPT models?

• BERT is primarily used for text generation, while GPT is used for text
classification.

• BERT is designed for unsupervised learning, while GPT is designed for

supervised learning.

• BERT is trained on a smaller dataset than GPT.

• BERT uses bidirectional context, while GPT uses unidirectional context.

• BERT uses a decoder-only architecture, while GPT uses an encoder-only

architecture.

Correct Answer: BERT uses bidirectional context, while GPT uses unidirectional
context.

Question #320: Which type of neural network is specifically designed for

processing sequential data by maintaining a 'memory' of previous inputs?

• Generative Adversarial Network (GAN)

• Feedforward Neural Network

• Self-Organizing Map (SOM)

• Convolutional Neural Network (CNN)

• Recurrent Neural Network (RNN)

Correct Answer: Recurrent Neural Network (RNN)

124 | P a g e
Question #321: You're designing a search engine and want to handle wildcard
queries of the form "pre*suf" (single * in the middle). You decide to use a
combination of B-trees. Which approach is MOST efficient and accurate?

• Use a permuterm index and look up 'suf$pre*'.

• Use a k-gram index with the Boolean query 'pre AND suf'.

• Use a single B-tree and traverse it twice, once for 'pre' and once for 'suf'.

• Use a regular B-tree for 'pre' and a reverse B-tree for 'suf', then intersect the
results.

• Use a single reverse B-tree and traverse it for 'fus*erp'.

Correct Answer: Use a regular B-tree for 'pre' and a reverse B-tree for 'suf', then
intersect the results.

Question #322: Level 3: Why is query optimization important in Information

Retrieval?

• To minimize the time taken to process a query

• To increase the number of documents retrieved

• To reduce the size of the inverted index

• To improve the user interface

• To ensure all relevant documents are found

Correct Answer: To minimize the time taken to process a query

Question #323: What is a "posting" in the context of an inverted index?

• A measure of relevance

• A document in the collection

• A query submitted by a user

125 | P a g e
 • An entry in a postings list, indicating a term's occurrence in a document

• A term in the dictionary

Correct Answer: An entry in a postings list, indicating a term's occurrence in a

document

Question #324: Which step involves determining the correct encoding and
decoding of a document's character sequence?

• Collecting documents

• Tokenizing text

• Indexing documents

• Linguistic preprocessing

• Decoding byte sequences into characters

Correct Answer: Decoding byte sequences into characters

Question #325: What is meant by "index granularity"?

• The number of indexes used

• The average length of documents

• The size of the units indexed (e.g., whole documents, paragraphs,

sentences)

• The size of the vocabulary

• The compression technique used

Correct Answer: The size of the units indexed (e.g., whole documents, paragraphs,
sentences)

Question #326: Question: What is the purpose of a compound splitter module in

retrieval systems for languages like German?

126 | P a g e
 • To identify the sentiment of compound words

• To translate compound words into English

• To split compound nouns into their constituent words

• To generate synonyms for compound words

• To correct spelling errors in compound words

Correct Answer: To split compound nouns into their constituent words

Question #327: What is the main advantage of using skip pointers in postings list
intersection?

• It makes it easier to update the index

• It allows for the use of wildcards in queries

• It reduces the number of comparisons needed, making intersection faster

• It improves the accuracy of stemming

• It eliminates the need for a dictionary

Correct Answer: It reduces the number of comparisons needed, making

intersection faster

Question #328: Question: What is asymmetric query expansion?

• Expanding the query with different terms depending on the original query
term

• Expanding the query with terms that are only relevant to a specific subset of
the documents in the collection

• Expanding the query with terms that are antonyms of the original query
terms

• Expanding the query with terms that are more general than the original query
terms

127 | P a g e
 • Expanding the query with terms that are phonetically similar to the original
query terms

Correct Answer: Expanding the query with different terms depending on the
original query term

Question #329: What is a common drawback of using the "AND" operator

extensively in Boolean queries?

• Inability to handle phrase queries

• Slow query processing

• High recall but low precision

• Difficulty in ranking results

• High precision but low recall

Correct Answer: High precision but low recall

Question #330: Level 5: A user is searching for information about "apple pie." The
system retrieves documents about Apple (the company) and pie charts. What is the
primary issue illustrated here?

• Lack of stop word removal

• Poor query formulation

• Lack of stemming

• Inefficient indexing

• Polysemy

Correct Answer: Polysemy

Question #331: What is "tokenization" in the context of IR?

• Ranking documents

128 | P a g e
 • Breaking a stream of text into words, phrases, symbols, or other meaningful
elements (tokens)

• Assigning weights to terms

• Performing stemming

• Creating an inverted index

Correct Answer: Breaking a stream of text into words, phrases, symbols, or other
meaningful elements (tokens)

Question #332: Which of the following best describes the purpose of allowing
asymmetric expansion of query terms?

• To allow partial matches without unintended broadening of queries.

• To provide flexibility in how terms can match while avoiding excessive query
expansion.

• To balance the trade-offs between space and processing costs for storage.

• To ensure that both methods of query expansion are less efficient than
equivalence classing.

• To handle different casings of a term effectively, such as matching

"Windows" but not "window".

Correct Answer: To handle different casings of a term effectively, such as matching

"Windows" but not "window".

Question #333: Question: What is the purpose of "query optimization" in Boolean

retrieval?

• To improve the visual presentation of search results

• To automatically correct spelling errors in the query

• To reduce the amount of work needed to answer a query

• To provide personalized search results based on user preferences

129 | P a g e
 • To translate the query into different languages

Correct Answer: To reduce the amount of work needed to answer a query

Question #334: What is "case-folding" in text normalization?

• Converting all letters to lowercase

• Removing punctuation

• Converting all letters to uppercase

• Removing diacritics

• Identifying proper nouns

Correct Answer: Converting all letters to lowercase

Question #335: What is a "proximity operator" in an extended Boolean model?

• An operator that finds the union of two postings lists

• An operator that expands a query with synonyms

• An operator that finds the intersection of two postings lists

• An operator specifying that terms must occur close to each other in a

document

• An operator that performs stemming

Correct Answer: An operator specifying that terms must occur close to each other
in a document

Question #336: What is the primary purpose of stemming?

• To reduce inflectional and sometimes derivational forms of a word to a

common base form

• To identify the root form of a word

• To expand the vocabulary

130 | P a g e
 • To find synonyms for a word

• To correct spelling errors

Correct Answer: To reduce inflectional and sometimes derivational forms of a word

to a common base form

Question #337: What information is stored in a positional index, in addition to

document IDs?

• The positions of the term's occurrences within each document

• The frequency of the term in the entire collection

• The grammatical role of the term

• Synonyms of the term

• The language of the document

Correct Answer: The positions of the term's occurrences within each document

Question #338: Question: What is the main challenge with using phrase indexes?

• They do not support Boolean queries

• They do not support ranked retrieval

• They are difficult to implement

• They require a large amount of storage space

• They are computationally expensive

Correct Answer: They require a large amount of storage space

Question #339: What constitutes an appropriate document unit for indexing in

cases where email messages contain multiple files or are split into separate pages?

• The entire book or collection as one document

• The entire folder as a single document

131 | P a g e
 • Each email message, including any attached files as separate documents

• Each individual sentence within a document

• Each paragraph within a document

Correct Answer: Each email message, including any attached files as separate
documents

Question #340: Why is the Jaccard coefficient, when used with a k-gram index for
spelling correction, more nuanced than simply counting the number of matching k-
grams?*

• The Jaccard coefficient works better for short queries.

• The Jaccard coefficient doesn't require a k-gram index.

• The Jaccard coefficient accounts for the lengths of the terms being
compared, reducing false positives.

• The Jaccard coefficient is faster to compute.

• The Jaccard coefficient is only used for context-sensitive correction.

Correct Answer: The Jaccard coefficient accounts for the lengths of the terms
being compared, reducing false positives.

Question #341: If you have a three-term query where each term has five alternative
corrections suggested by isolated-term correction, how many possible corrected
phrases must be considered if you do not trim the space of corrected phrases but
instead try all six variants for each of the terms?

• 30

• 15

• 20

• 10

• 25

132 | P a g e
Correct Answer: 25

Question #342: In Westlaw, what does the operator "/s" signify?

• Terms must appear in the same paragraph

• Terms must appear in the same sentence

• Terms must be stemmed

• Terms must appear in the same document

• Terms must appear within a specified number of words

Correct Answer: Terms must appear in the same sentence

Question #343: Level 5: How might the presence of synonyms and polysemy
(words with multiple meanings) affect the performance of a retrieval system?

• They have no effect.

• They improve performance by increasing the number of matches.

• They can lead to both false positives and false negatives.

• They only affect the Boolean Model.

• They only affect the Vector Space Model.

Correct Answer: They can lead to both false positives and false negatives.

Question #344: In the Boolean retrieval model, how is the query "A and B"
processed using an inverted index?

• Retrieve all documents, then filter for A and B

• Retrieve postings lists for A and B, then intersect them

• Retrieve the postings list for B, then filter for documents containing A

• Retrieve the postings list for A, then filter for documents containing B

• Retrieve postings lists for A and B, then take their union

133 | P a g e
Correct Answer: Retrieve postings lists for A and B, then intersect them

Question #345: What is the primary purpose of "term normalization" in information

retrieval?

• To handle spelling errors in queries

• To reduce the size of the index

• To assign weights to terms based on their importance

• To improve the speed of query processing

• To group together different forms of the same word (e.g., stemming,

lemmatization)

Correct Answer: To group together different forms of the same word (e.g.,
stemming, lemmatization)

Question #346: When determining the appropriate size of document units in an

information retrieval system, what are the primary trade-offs regarding precision
and recall?

• There is no trade-off; both precision and recall can be maximized

simultaneously.

• Larger document units reduce false positives but may miss important
passages.

• Decreasing unit size enhances recall without affecting precision.

• Increasing unit size improves precision but sacrifices recall.

• Smaller document units improve recall but increase the likelihood of

spurious matches.

Correct Answer: Larger document units reduce false positives but may miss
important passages.

134 | P a g e
Question #347: Why is a term-document incidence matrix generally a sparse
matrix?

• Most terms appear in only a small fraction of the documents

• The matrix is used only for small collections

• The matrix is stored in a compressed format

• Both A and B

• Most documents contain only a small fraction of the total vocabulary terms

Correct Answer: Most documents contain only a small fraction of the total
vocabulary terms

Question #348: Difficulty: level1 Question: What is the purpose of "Normalization"

in Information Retrieval?

• To identify the language of a document.

• To remove punctuation marks.

• To convert all words to lowercase.

• To create equivalence classes of terms.

• To reduce words to their root form

Correct Answer: To create equivalence classes of terms.

Question #349: A query is "life insurance company employee". Using the extended
biword indexing model (NX*N), which of the following Boolean queries would be
generated? (Assume part-of-speech tagging is perfect).

• "life insurance" AND "insurance company employee"

• "life insurance" AND "insurance company" AND "company employee"

• life insurance company employee

• "life insurance" AND "company employee"

135 | P a g e
 • "life insurance company" AND "company employee"

Correct Answer: "life insurance company" AND "company employee"

Question #350: You have a query "flew form Heathrow" which returns very few
results. You suspect a spelling error. You've generated the following potential
corrections for each term:

• flew: flew, flew, flue

• form: form, from, fore

• Heathrow: Heathrow, Heathraw Which of the following strategies is the

MOST computationally efficient AND likely to yield the best context-sensitive
correction, assuming you have access to term and bi-gram frequencies from
a large corpus?

• Generate all 27 possible corrected phrases and run each as a query,

selecting the one with the most results.

• Use the isolated-term corrections and bi-gram frequencies to prune the

search space. Prioritize extending corrections with high bi-gram frequency
(e.g., 'flew from') and only consider 'Heathrow' corrections if the top bi-gram
corrections yield few results.

• Calculate the edit distance between the original query and all 27 (3x3x3)
possible corrected phrases.

• Generate all 27 possible corrected phrases, calculate the probability of each

phrase using a language model, and select the most probable.

• Calculate the Jaccard coefficient between the original query's bigrams and
the bigrams of all 27 possible corrected phrases.

Correct Answer: Use the isolated-term corrections and bi-gram frequencies to

prune the search space. Prioritize extending corrections with high bi-gram
frequency (e.g., 'flew from') and only consider 'Heathrow' corrections if the top bi-
gram corrections yield few results.

136 | P a g e
Question #351: What is a potential problem with using a very large stop list?

• It can remove terms that are important for certain queries, harming recall

• It can make the index too large

• It can slow down query processing

• It can prevent the use of positional indexes

• It can make stemming less effective

Correct Answer: It can remove terms that are important for certain queries,
harming recall

Question #352: Which of the following sentences would falsely match the query
"mon*h" if the search were to simply use a conjunction of bigrams?

• mon*h is incorrect.

• mont*her is incorrect.

• mon*h matches correctly.

• This is not relevant.

• mont*her is correct.

Correct Answer: mont*her is incorrect.

Question #353: What is a key difference between a forward index and an inverted
index?

• A forward index is more space-efficient

• A forward index is faster for query processing

• A forward index stores documents, and an inverted index stores terms

• A forward index can handle phrase queries, while an inverted index cannot

• A forward index is used for Boolean retrieval, and an inverted index is used
for ranked retrieval

137 | P a g e
Correct Answer: A forward index is used for Boolean retrieval, and an inverted
index is used for ranked retrieval

Question #354: Difficulty: 1 Question 1: What is the primary function of the

"Merge" algorithm in Boolean retrieval (as described in slides 3-5)?

• To create new postings lists

• To locate terms in the dictionary

• To retrieve postings lists

• To intersect the document sets of two postings lists

• To sort documents alphabetically

Correct Answer: To intersect the document sets of two postings lists

Question #355: What is the main advantage of using a skip list in a postings list?

• It supports proximity searches

• It eliminates the need for stemming

• It allows for faster intersection of postings lists

• It reduces the size of the postings list

• It enables phrase queries

Correct Answer: It allows for faster intersection of postings lists

Question #356: Question: Which data structure is most suitable for storing
postings lists to allow cheap insertion of documents into postings lists following
updates?

• Fixed length array

• Hash Table

• Balanced Tree

138 | P a g e
 • Singly linked list

• Variable length array

Correct Answer: Singly linked list

Question #357: Level 2: What is a "posting" in the context of an Inverted Index?

• A document ID where a term appears

• A list of all terms in a document

• The total number of documents in the collection

• A single term in the dictionary

• The frequency of a term in a document

Correct Answer: A document ID where a term appears

Question #358: A user types the query "Romeo and Juliet are from Verona" into a
search engine. How does the search engine process this phrase to support the
query efficiently?

• The search engine splits the input into individual words:

• The search engine applies Boolean logic to match the query as a single
phrase without splitting into biwords.

• The search engine recognizes the phrase as two extended biwords:

• The search engine converts consecutive terms into biwords, treating them as
vocabulary terms.

• The search engine identifies and treats

Correct Answer: The search engine converts consecutive terms into biwords,
treating them as vocabulary terms.

Question #359: What is the main advantage of using skip pointers in a postings list?

139 | P a g e
 • To support wildcard queries

• To enable efficient phrase queries

• To skip over documents that do not contain all query terms, thus speeding
up intersection operations

• To allow for faster merging of postings lists

• To reduce the storage space required for the postings list

Correct Answer: To skip over documents that do not contain all query terms, thus
speeding up intersection operations

Question #360: Which data structure is particularly useful for handling trailing
wildcard queries efficiently?

• B-tree

• Hash Table

• Soundex Index

• k-gram index

• Permuterm Index

Correct Answer: Permuterm Index

Question #361: In query optimization, why is processing terms in order of

increasing document frequency generally preferred?**

• It guarantees the most relevant documents are found first

• It allows for parallel processing of terms

• It minimizes the size of intermediate result sets, improving efficiency

• It's easier to implement

• It works best with the Boolean model

140 | P a g e
Correct Answer: It minimizes the size of intermediate result sets, improving
efficiency

Question #362: In a positional index, what does the "position" information

represent?

• The paragraph number where the term appears

• The sentence number where the term appears

• The line number where the term appears

• The offset of the term from the beginning of the document

• The page number where the term appears

Correct Answer: The offset of the term from the beginning of the document

Question #363: Level 1: Which of the following is NOT a typical component of an

Information Retrieval system?

• Database Management System (DBMS)

• User Interface

• Indexer

• Query Processor

• Retrieval Model

Correct Answer: Database Management System (DBMS)

Question #364: When implementing a spelling correction system using a

permuterm index for efficient retrieval, what is an effective approach to reduce both
errors and computational overhead?

• Applying a term-document incidence matrix for dynamic updates and

efficient lookups.

141 | P a g e
 • Combining initial-letter matching with a refined rotation strategy that omits a
suffix before traversal.

• Employing vector space models to map terms into a continuous space for
similarity retrieval.

• Utilizing all possible rotations of the query string without omitting any
suffixes.

• Restricting search to terms starting with exactly one specific letter of the
query.

Correct Answer: Combining initial-letter matching with a refined rotation strategy

that omits a suffix before traversal.

Question #365: When determining the appropriate size of document units in an

information retrieval system, what is the primary trade-off regarding precision and
recall?

• Increasing the granularity of document units leads to better recall but

reduced precision.

• Decreasing the granularity of document units increases precision at the cost

of recall.

• The trade-off between precision and recall depends on the specific needs of
the users and the document collection.

• Increasing the granularity of document units improves both precision and

recall.

• Decreasing the granularity of document units decreases precision but

improves recall.

Correct Answer: Increasing the granularity of document units leads to better recall
but reduced precision.

142 | P a g e
Question #366: Referring to the exercise on slide 15, how could the frequency of
"countrymen" be used to evaluate the best query evaluation order for friends AND
romans AND (NOT countrymen)?**

• Estimate the size of NOT countrymen by using the total number of

documents minus the frequency of countrymen.

• Ignore 'countrymen' since it's negated.

• Process 'friends' and 'romans' first, then apply the negation of 'countrymen'
to the result.

• Process 'countrymen' first, then 'friends' and 'romans'.

• The frequency of countrymen is irrelevant

Correct Answer: Estimate the size of NOT countrymen by using the total number of
documents minus the frequency of countrymen.

Question #367: What is the primary data structure used to determine if a query
term exists in the vocabulary of an inverted index?

• Trie

• Postings List

• Hash Table

• Binary Search Tree

• Dictionary

Correct Answer: Dictionary

Question #368: What is the optimal strategy for processing terms in a Boolean
query to minimize the total amount of work?

• Alternate between high and low document frequency terms

• Process terms in order of decreasing document frequency

• Process terms in arbitrary order

143 | P a g e
 • Process terms in order of increasing document frequency

• Always process the term with the largest postings list first

Correct Answer: Process terms in order of increasing document frequency

Question #369: Consider the query: (A OR B) AND (C OR D). Assume we have the
following document frequencies: df(A) = 10, df(B) = 20, df(C) = 5, df(D) = 30. Which
of the following processing orders is likely to be MOST efficient, and why?

• Process A, then B, then C, then D individually, and then combine

• A and C first, then B and D.

• It doesn't matter; all orders are equally efficient.

• (C OR D) first, then (A OR B)

• (A OR B) first, then (C OR D)

Correct Answer: (C OR D) first, then (A OR B)

Question #370: What is the main advantage of Boolean retrieval models,

particularly for professional searchers?

• Automatic query expansion

• Ease of use for novice users

• Automatic ranking of results

• Precision and control over the retrieved results

• High recall

Correct Answer: Precision and control over the retrieved results

Question #371: Question: In the context of information retrieval, what is

considered a term when discussing classification and clustering in Chapters 13–18
of the text?

144 | P a g e
 • Language identification uses short character subsequences but doesn't
affect tokenization rules.

• Specific tokens like email addresses should always be treated as separate

terms.

• Omitting certain tokens from indexing can still allow effective search
capabilities.

• A term is defined as a normalized word that must always be included in the

dictionary.

• In classification and clustering contexts, a term does not need to be

included in the dictionary.

Correct Answer: In classification and clustering contexts, a term does not need to
be included in the dictionary.

Question #372: What is the purpose of the Term-Document Incidence Matrix?

• To store the frequency of each term in each document

• To list all the terms in the vocabulary

• To calculate the similarity between documents

• To represent which documents contain which terms (using 0s and 1s)

• To store the positions of terms within documents

Correct Answer: To represent which documents contain which terms (using 0s and
1s)

Question #373: Difficulty: 2 Question 4: What is a major drawback of using only

biword indexes for phrase queries (slide 6)?

• They cannot handle phrase queries at all.

• They are too slow for practical use.

• They only work for two-word phrases.

145 | P a g e
 • They require proximity operators.

• They can lead to false positives.

Correct Answer: They can lead to false positives.

Question #374: Why is normalization (equivalence classing of terms) important in

Information Retrieval?

• It identifies the language of the documents

• It speeds up the stemming process

• It corrects grammatical errors

• It reduces the size of the documents

• It helps match different forms of the same word (e.g., "USA" and "U.S.A.")

Correct Answer: It helps match different forms of the same word (e.g., "USA" and
"U.S.A.")

Question #375: You have a document collection where many documents contain
lists of names and associated information (e.g., employee directories). You want to
allow users to search for people by name and job title, even if the title appears
several words away from the name. Which indexing strategy would be MOST
appropriate, and why would other strategies be less effective?

• A simple term frequency index, because it's efficient.

• Positional index, because it allows for proximity searches.

• Biword index, because names and titles are often two words.

• Boolean index with stop word removal, because it's the simplest.

• Phrase index, because it handles multi-word units.

Correct Answer: Positional index, because it allows for proximity searches.

146 | P a g e
Question #376: In context-sensitive spelling correction, why is it often necessary to
limit the number of alternative query phrases considered?

• Because the edit distance calculation is less accurate for longer phrases.

• To prioritize corrections that are phonetically similar to the original query.

• To avoid suggesting corrections that are too dissimilar to the user's intent.

• Because the number of possible combinations of term corrections can grow

very large, making exhaustive search impractical.

• To reduce the computational cost of calculating edit distances.

Correct Answer: Because the number of possible combinations of term

corrections can grow very large, making exhaustive search impractical.

Question #377: What is "stemming" in the context of IR?

• Removing stop words

• Finding synonyms for words

• Tokenizing text

• Decoding character encodings

• Reducing words to their root or base form (e.g., "running" -> "run")

Correct Answer: Reducing words to their root or base form (e.g., "running" -> "run")

Question #378: Level 4: How does the concept of "term frequency" (tf) contribute
to document ranking in many retrieval models?

• tf is only used in the Boolean Model.

• tf is irrelevant for ranking.

• Higher tf always means higher relevance.

• tf is used to normalize document length.

147 | P a g e
 • Higher tf generally indicates a stronger relationship between the term and
the document.

Correct Answer: Higher tf generally indicates a stronger relationship between the

term and the document.

Question #379: What is a potential trade-off when determining the granularity of

document units in an information retrieval system?

• Larger document units reduce the need for semantic processing

• Precision may increase but recall might decrease

• Tokenization becomes unnecessary when using large document units

• Granularity has no impact on search effectiveness

• Smaller document units make indexing more efficient

Correct Answer: Precision may increase but recall might decrease

Question #380: What data structure is most suitable for a postings list in an
inverted index?

• Variable length array

• Binary search tree

• Singly linked list

• Hash table

• Fixed-length array

Correct Answer: Singly linked list

Question #381: Which retrieval model treats documents and queries as sets of
terms and uses set operations (AND, OR, NOT) to find matching documents?

• Language Model

148 | P a g e
 • Inferred Model

• Vector Space Model

• Probabilistic Retrieval Model

• Boolean Retrieval Model

Correct Answer: Boolean Retrieval Model

Question #382: When indexing a document containing text in multiple languages,

what is the recommended approach for ensuring accurate retrieval and
consistency?

• Identify the language used in each segment of the document and apply
language-specific tokenization and normalization rules.

• Ignore language differences and treat all text as a single language for
indexing purposes.

• Tokenize the entire document using one writing system and apply a single set
of normalization rules.

• Apply default heuristics without specific language-based rules.

• Use a global Soundex algorithm to convert all text into a common phonetic
form.

Correct Answer: Identify the language used in each segment of the document and
apply language-specific tokenization and normalization rules.

Question #383: You are correcting the misspelled query "appple". You are using a
combination of a bigram index and edit distance. You've identified "apple" and
"apply" as potential corrections based on bigram overlap. Which of the following
statements is MOST accurate in guiding your next step?

• You should reject both 'apple' and 'apply' because they don't share all
bigrams with 'appple'.

149 | P a g e
 • You should calculate the edit distance to both 'apple' and 'apply' and
consider additional factors like term frequency.

• You should only consider 'apple' because it shares more bigrams with
'appple'.

• 'apply' is definitively the correct spelling because it has a lower edit

distance.

• 'apple' is definitively the correct spelling because it has a lower edit

distance.

Correct Answer: You should calculate the edit distance to both 'apple' and 'apply'
and consider additional factors like term frequency.

Question #384: In the vector space model, what is the purpose of length
normalization?

• To ensure that all vectors have the same length.

• To reduce the impact of frequent terms.

• To increase the impact of rare terms.

• To penalize longer documents.

• To give more weight to shorter documents.

Correct Answer: To ensure that all vectors have the same length.

Question #385: What is the purpose of using log-frequency weighting in

information retrieval?

• To normalize the length of documents.

• To completely eliminate the impact of term frequency.

• To reduce the impact of very high term frequencies, providing a more

balanced score.

• To convert term frequencies into binary values.

150 | P a g e
 • To increase the impact of rare terms in a document.

Correct Answer: To reduce the impact of very high term frequencies, providing a
more balanced score.

Question #386: What is the purpose of inexact top-K retrieval methods?

• To reduce the computational cost of finding the top K documents

• To index documents faster

• To guarantee that the top K documents are always retrieved accurately

• To handle Boolean queries more efficiently

• To improve the precision of the search results

Correct Answer: To reduce the computational cost of finding the top K documents

Question #387: In the context of information retrieval evaluation, which scenario

best illustrates the trade-off between precision and recall, and how might a system
designer address this trade-off based on the specific application?

• A system that indexes all words versus a system that uses a stop list.

• A system that uses a large index versus a system that uses a small index.

• A system that uses tf-idf weighting versus a system that uses binary term
weighting.

• A system designed for legal discovery where missing a single relevant

document could have severe consequences versus a web search engine
where users are primarily concerned with the quality of the first page of
results.

• A system that uses stemming versus a system that does not use stemming.

Correct Answer: A system designed for legal discovery where missing a single
relevant document could have severe consequences versus a web search engine
where users are primarily concerned with the quality of the first page of results.

151 | P a g e
Question #388: Which of the following is a disadvantage of using the "Bag of
Words" model in information retrieval?

• It ignores the order and context of words in a document.

• It is difficult to calculate term frequencies.

• It requires a large amount of storage space.

• It is computationally expensive to implement.

• It cannot be used with stemming or lemmatization.

Correct Answer: It ignores the order and context of words in a document.

Question #389: Difficulty: level1 Question: What is the primary goal of the
tokenization process in Information Retrieval?

• To break down text into individual terms

• To create an inverted index

• To calculate term frequencies

• To perform stemming

• To identify stop words

Correct Answer: To break down text into individual terms

Question #390: What is the "term-document incidence matrix"?

• A tool for query optimization

• A matrix showing the presence or absence of terms in documents

• A method for ranking documents

• A data structure for storing postings lists

• A list of all terms in a collection

Correct Answer: A matrix showing the presence or absence of terms in documents

152 | P a g e
Question #391: Level 3: What is a potential drawback of using only the Boolean
Retrieval Model?

• It does not support ranked retrieval

• It cannot handle large document collections

• It is difficult to implement

• It is too slow for interactive use

• It requires complex indexing techniques

Correct Answer: It does not support ranked retrieval

Question #392: What is the main disadvantage of using a Permuterm index for
wildcard queries?

• It is slow for queries with wildcards at the beginning of a term

• It is not suitable for large document collections

• It requires a complex merging process

• It cannot handle queries with multiple wildcards

• It significantly increases the size of the dictionary

Correct Answer: It significantly increases the size of the dictionary

Question #393: What is "query optimization" in the context of Boolean retrieval?

• Rewriting a query in disjunctive normal form

• Selecting the most efficient order to process terms and operations in a query

• Finding the most relevant documents for a query

• Expanding a query with synonyms

• Improving the precision and recall of a query

153 | P a g e
Correct Answer: Selecting the most efficient order to process terms and
operations in a query

Question #394: When a character is placed on the left side of its square box in a
text, what strategy helps ensure that all possible forms of a term are indexed
correctly?

• Users should enter hyphens whenever possible to enable generalization.

• Users should not use any special characters in their queries.

• The system should automatically convert all terms into one-word forms.

• Users should avoid entering hyphens.

• Indexing should be based on single words only.

Correct Answer: Users should enter hyphens whenever possible to enable

generalization.

Question #395: Why are skip pointers added to postings lists?

• To store positional information

• To speed up the intersection of postings lists by skipping over blocks of non-

matching documents

• To store term frequency information

• To handle phrase queries

• To compress the postings lists

Correct Answer: To speed up the intersection of postings lists by skipping over

blocks of non-matching documents

Question #396: What is a potential downside of using asymmetric expansion for

term normalization?

• It can lead to less efficient query processing.

154 | P a g e
 • It can make the index larger.

• It can result in fewer relevant documents being retrieved.

• It is not suitable for languages with complex morphology.

• It always improves retrieval accuracy.

Correct Answer: It can lead to less efficient query processing.

Question #397: What is the difference between stemming and lemmatization?

• Lemmatization always returns a valid word, while stemming may not

• Stemming uses a vocabulary, while lemmatization does not

• Stemming is more accurate than lemmatization

• Lemmatization removes derivational affixes, while stemming only removes

inflectional endings

• Stemming is only used for English, while lemmatization can be used for any
language

Correct Answer: Lemmatization always returns a valid word, while stemming may
not

Question #398: Level 1: In the Boolean Retrieval Model, what does the operator
"AND" signify between two terms?

• Either term can be present

• Only the second term must be present

• Both terms must be present

• Only the first term must be present

• Neither term should be present

Correct Answer: Both terms must be present

155 | P a g e
Question #399: Level 5: Why is the "ideal" query processing order not always
achievable in practice?

• Users often change their queries.

• The cost of merging postings lists can be unpredictable.

• The system might not have enough memory.

• Document frequencies are not always known in advance.

• Postings lists are not always sorted.

Correct Answer: The cost of merging postings lists can be unpredictable.

Question #400: When computing the Jaccard coefficient between a query q and a
term t during a postings scan, why do we not need to enumerate all k-grams from t?

• We need the exact string representation of t to compute Jaccard correctly.

• The term's length allows us to estimate the number of matching k-grams

accurately.

• Enumerating all k-grams from t would be too slow and unnecessary.

• The frequency of each k-gram in t is required for the calculation.

• The length of t is sufficient for accurate computation.

Correct Answer: Enumerating all k-grams from t would be too slow and
unnecessary.

Question #401: What is a potential disadvantage of using stop lists in information

retrieval systems?

• They can lead to loss of meaning in phrase queries or specific searches.

• They improve search precision by removing semantically nonselective

words.

• They eliminate the need for semantic content analysis.

156 | P a g e
 • They are universally used by all modern web search engines.

• They significantly increase the storage requirements for indexing.

Correct Answer: They can lead to loss of meaning in phrase queries or specific
searches.

Question #402: Which of the following tokenization approaches would NOT cause
problems in a search engine?

• Ignoring hyphens

• Separating numbers and URLs as metadata

• Using a phrase index for hyphenated terms

• Adding spaces within words

• Splitting on white space

Correct Answer: Splitting on white space

Question #403: Level 3: What is the role of stop words in Information Retrieval?

• They are frequently occurring, often insignificant words that are sometimes
removed.

• They are the most important words in a query.

• They are used to build the inverted index.

• They are words that are misspelled.

• They are keywords used to categorize documents.

Correct Answer: They are frequently occurring, often insignificant words that are
sometimes removed.

Question #404: Which of the following is the basis for ranked retrieval?

• Randomly ordering the documents.

157 | P a g e
 • Retrieving documents that contain all the query terms.

• Filtering out stop words from the query.

• Retrieving documents that contain at least one of the query terms.

• Assigning a score to each document based on its relevance to the query.

Correct Answer: Assigning a score to each document based on its relevance to the
query.

Question #405: What does the Jaccard coefficient measure?

• The similarity between two sets.

• The frequency of a term in a document.

• The inverse document frequency.

• The number of common terms between two documents.

• The distance between two documents.

Correct Answer: The similarity between two sets.

Question #406: What term refers to the process in IR systems where different token
forms are considered equivalent, allowing queries to match documents with
variations like "USA" and "U.S.A."?

• Stop words inclusion

• Query expansion

• Term normalization

• Token normalization

• Equivalence classing

Correct Answer: Term normalization

158 | P a g e
Question #407: Level 4 (Hard) 14. What is a key difference between isolated-
term and context-sensitive spelling correction?

• Isolated-term correction is more accurate

• Isolated-term correction uses edit distance, while context-sensitive uses k-

grams

• Context-sensitive correction considers the surrounding words in the query

• Isolated-term correction is faster

• Context-sensitive correction only works for single-word queries

Correct Answer: Context-sensitive correction considers the surrounding words in

the query.

Question #408: You are evaluating two information retrieval systems. System A has
a higher precision but lower recall, while System B has a lower precision but higher
recall. If the user values both precision and recall equally, which single metric
would be most appropriate to compare the two systems?

• Accuracy

• Precision

• F1-score

• Error Rate

• Recall

Correct Answer: F1-score

Question #409: Why does cos(SaS, PaP) exceed cos(SaS, WH)? Options:

• Because of differences in how normalization is applied between schemes.

• Because SaS and PaP are closer to each other compared to SaS and WH.

• Because WH uses a different length normalization method.

159 | P a g e
 • Because PaP applies idf weighting while WH does not.

• Because the term weights under PaP are generally higher than those under
WH for matching terms.

Correct Answer: Because PaP applies idf weighting while WH does not.

Question #410: What does "explicit politeness" refer to in web crawling?

• Avoiding crawling during peak traffic hours.

• Prioritizing websites with high page rank.

• Specifications from webmasters on what portions of a site can be crawled,

often defined in [Link].

• Crawling only websites that explicitly grant permission.

• Maintaining a consistent crawl rate across all websites.

Correct Answer: Specifications from webmasters on what portions of a site can be

crawled, often defined in [Link].

Question #411: A web crawler discovers two URLs that point to virtually identical
content but have different URL structures and are hosted on different domains.
Which strategy would be most effective in avoiding duplicate indexing while
preserving potentially valuable information?

• Indexing both URLs separately without further analysis.

• Using content-based deduplication techniques to identify and merge the

content under a single canonical URL.

• Ignoring the second URL encountered.

• Redirecting users from the second URL to the first URL.

• Penalizing both domains in search rankings for hosting duplicate content.

Correct Answer: Using content-based deduplication techniques to identify and

merge the content under a single canonical URL.

160 | P a g e
Question #412: In the context of tf-idf weighting, how does inverse document
frequency (IDF) affect the weight of a term?

• It decreases the weight of terms that appear in many documents.

• It normalizes the term frequency.

• It has no effect on the weight of a term.

• It only affects the weight of terms in the query, not in the documents.

• It increases the weight of terms that appear in many documents.

Correct Answer: It decreases the weight of terms that appear in many documents.

Question #413: In ranked retrieval, what is the primary purpose of assigning a score
to each document in the collection with respect to a query?

• To identify the language of the document.

• To measure the length of the document.

• To categorize the documents into predefined topics.

• To determine if the document is relevant or not.

• To rank-order the documents based on their relevance to the query.

Correct Answer: To rank-order the documents based on their relevance to the

query.

Question #414: In evaluating information retrieval systems, what does Mean

Average Precision (MAP) primarily measure?

• The average number of relevant documents retrieved per query.

• The precision and recall at different recall levels.

• The average of the average precisions for each query.

• The average precision at a fixed rank K across all queries.

161 | P a g e
 • The average precision of the first relevant document retrieved.

Correct Answer: The average of the average precisions for each query.

Question #415: What is the main purpose of using the F measure in information
retrieval evaluation?

• To reward systems that retrieve relevant documents quickly.

• To penalize systems that retrieve too many documents.

• To combine precision and recall into a single metric.

• To measure the size of the index.

• To measure the speed of the retrieval system.

Correct Answer: To combine precision and recall into a single metric.

Question #416: Consider two documents: Document A contains the word "apple" 5
times, and Document B contains the word "apple" 10 times. Without considering
any other factors, which document would be ranked higher if using only term
frequency (tf) as the ranking factor?

• They would be ranked the same.

• It depends on the inverse document frequency of "apple".

• It depends on the length of the documents.

• Document A.

• Document B.

Correct Answer: Document B.

Question #417: What is the primary goal of ranked retrieval in information retrieval?

• To order documents by their relevance to a query.

• To retrieve documents in alphabetical order.

162 | P a g e
 • To retrieve only the most relevant document.

• To retrieve all documents in the collection.

• To retrieve documents randomly.

Correct Answer: To order documents by their relevance to a query.

Question #418: R-precision adjusts for what factor when evaluating ranked retrieval
results?

• The number of documents retrieved.

• The size of the document collection.

• The number of relevant documents for a query.

• The length of the query.

• The speed of the retrieval system.

Correct Answer: The number of relevant documents for a query.

Question #419: In the context of scoring for ranked retrieval, what does a score of 0
typically indicate?

• The document is highly relevant to the query.

• The document contains all the query terms.

• The document is not relevant to the query.

• The document's relevance is unknown.

• The document is the most relevant in the collection.

Correct Answer: The document is not relevant to the query.

Question #420: Question: Which of the following is the biggest problem that ranked
retrieval solves compared to boolean retrieval?

• Lack of support for stemming and lemmatization.

163 | P a g e
 • Difficulty in understanding complex query languages.

• Returning too few or too many results.

• Inability to handle large document collections.

• Difficulty in assigning scores to documents.

Correct Answer: Returning too few or too many results.

Question #421: In the context of information retrieval, what is the "F1-score"?

• The number of documents in the collection.

• A measure of the storage space used by the index.

• The harmonic mean of precision and recall.

• A measure of the speed of the retrieval system.

• The sum of precision and recall.

Correct Answer: The harmonic mean of precision and recall.

Question #422: Question: In determining the correct tokenization for names like
"O’Neil," what is the correct tokenization approach?

• o neill

• o’neill

• o’ neill

• neill

• oneill

Correct Answer: o’neill

Question #423: In information retrieval, how are general wildcard queries handled
when a search string contains a single '*' symbol not confined to either the
beginning or end of the string?

164 | P a g e
 • Leading wildcard query, employing only a reverse B-tree.

• Using two B-trees: one normal and one reverse, then taking the intersection
of the results.

• Combining a prefix tree with an inverted index for efficient lookups.

• Trailing wildcard query only, using a standard inverted index.

• Building a single B-tree with a special structure to support any kind of

wildcard query.

Correct Answer: Using two B-trees: one normal and one reverse, then taking the
intersection of the results.

Question #424: During spelling correction, which method is used to retrieve

vocabulary terms that have many k-grams in common with the query?

• Relying only on edit distances without incorporating any k-gram analysis.

• Employing a fixed k-value without considering sufficient matches from the

query's bigrams.

• Using a 2-gram index to scan and find terms with at least two bigrams
matching the query.

• Not utilizing k-gram indexes for spelling correction, instead relying solely on
edit distance.

• Applying the Jaccard coefficient to measure overlap between the query's

and term's k-grams during retrieval.

Correct Answer: Using a 2-gram index to scan and find terms with at least two
bigrams matching the query.

Question #425: What is the purpose of an "inverted index" in IR?

• To perform Boolean operations

• To map terms to the documents where they occur

165 | P a g e
 • To rank documents based on relevance

• To optimize query processing speed

• To store documents in a structured format

Correct Answer: To map terms to the documents where they occur

Question #426: What is "lemmatization"?

• Chopping off the ends of words

• Creating a biword index

• Finding synonyms

• Reducing inflectional forms of a word to its base or dictionary form (lemma)

• Removing stop words

Correct Answer: Reducing inflectional forms of a word to its base or dictionary

form (lemma)

Question #427: What is tokenization in the context of Information Retrieval?

• Identifying the language of a document

• Removing punctuation

• Separating words with spaces

• Converting all text to lowercase

• Chopping a character sequence into pieces called tokens

Correct Answer: Chopping a character sequence into pieces called tokens

Question #428: Which of the following strategies is recommended for handling

diacritics in search engine normalization?

• Using truecasing for accurate capitalization decisions

• Removing diacritics because users often omit them when searching

166 | P a g e
 • Equating British and American spellings (e.g., colour/color)

• Keeping all original characters to preserve distinctions

• Converting all letters to lowercase to handle capitalization variations

Correct Answer: Removing diacritics because users often omit them when
searching

Question #429: You are comparing two stemming algorithms, Stemmer A and
Stemmer B. On a test set of 100 queries, Stemmer A improves precision on 20
queries, has no effect on 60 queries, and reduces precision on 20 queries. Stemmer
B improves precision on 30 queries, has no effect on 40 queries, and reduces
precision on 30 queries. Which stemmer is better overall, and why is this a
potentially misleading metric?

• They are equally good because they have the same number of positive and
negative effects.

• Stemmer A is better because it has fewer negative effects.

• Both are bad because they don't improve the majority of queries.

• Stemmer B is better because it has more positive effects.

• It's impossible to say which is better without knowing the magnitude of the
precision changes.

Correct Answer: It's impossible to say which is better without knowing the
magnitude of the precision changes.

Question #430: Which statement correctly describes a way that asymmetric

expansion of query terms can model users' expectations?

• It relies solely on normalizing tokens to remove diacritics and case

differences.

• It requires more storage space but offers no flexibility in query matching.

167 | P a g e
 • It is less efficient than equivalence classing due to increased processing at
query time.

• It allows for the use of equivalence classes, ensuring identical expansions

for all terms.

• It ensures that searches for "Windows" match documents about the

operating system but not about physical windows.

Correct Answer: It ensures that searches for "Windows" match documents about
the operating system but not about physical windows.

Question #431: Level 1: What is the primary function of "Tokenization" in text

processing?

• Combining words into phrases

• Removing punctuation

• Sorting words alphabetically

• Splitting text into individual words or terms

• Converting text to lowercase

Correct Answer: Splitting text into individual words or terms

Question #432: In the Vector Space Model, how is the similarity between a
document and a query typically calculated?

• Using the Jaccard coefficient of the term sets

• Using the cosine similarity between the vectors

• Using the number of shared terms

• Using the Euclidean distance between the vectors

• Using the Boolean AND operation

Correct Answer: Using the cosine similarity between the vectors

168 | P a g e
Question #433: In the context of dynamic indexing, what is an "auxiliary index"?

• An index used for wildcard queries

• An index used for spell correction

• An index used to store deleted documents

• A small index held in memory to store new or updated documents

• An index that stores term frequencies

Correct Answer: A small index held in memory to store new or updated documents

Question #434: What is a limitation of using a biword index for phrase queries?

• It cannot handle proximity queries

• It can lead to a very large dictionary size

• It is slower than a positional index

• It requires stemming

• It cannot handle Boolean operators

Correct Answer: It can lead to a very large dictionary size

Question #435: What is an advantage of using a positional index, relative to bi-word

index?

• It stores term frequencies

• Can support proximity queries and is more flexible in the phrases supported.

• It uses a term-document incidence matrix

• It uses skip pointers

• It performs stemming

Correct Answer: Can support proximity queries and is more flexible in the phrases
supported.
169 | P a g e
Question #436: Level 2: What is the main advantage of using an Inverted Index over
a Term-Document Incidence Matrix for large collections?

• Simpler to implement

• Easier to update

• Supports ranked retrieval directly

• More efficient storage and retrieval

• Better for complex queries

Correct Answer: More efficient storage and retrieval

Question #437: Which of the following illustrates a problem caused by splitting

tokens on spaces?

• San Francisco-Los Angeles

• Co-education

• Hewlett-Packard

• [Link]

• Mar 11, 1983

Correct Answer: San Francisco-Los Angeles

Question #438: When a search engine processes a Boolean combination of

wildcard queries like red AND feri, what is the appropriate semantic interpretation?

• Intersection

• OR of ANDs

• OR of ANDs

• XOR

• AND of ORs

170 | P a g e
Correct Answer: AND of ORs

Question #439: Difficulty: 2 Question 3: According to slide 11, what is the primary
goal of query optimization in a Boolean retrieval system?

• To reduce the processing time of a query

• To make queries look more complex

• To use more proximity operators

• To find the most relevant documents

• To increase the number of results returned

Correct Answer: To reduce the processing time of a query

Question #440: Level 2: Which of the following best describes "Normalization" in

text processing?

• Splitting text into tokens

• Converting words to a common form (e.g., lowercase)

• Identifying the root form of words

• Removing stop words

• Calculating term weights

Correct Answer: Converting words to a common form (e.g., lowercase)

Question #441: How many of the five possible tokenizations of Mr. O’Neill’s name
would result in a match when querying for "o’neill and capital"?

• 1

• 5

• 2

• 4

171 | P a g e
 • 3

Correct Answer: 3

Question #442: What is "linguistic preprocessing" in IR?

• Creating the dictionary of an inverted index

• Merging postings lists

• Applying operations like stemming and stop word removal to tokens

• Sorting postings lists

• Evaluating the effectiveness of a retrieval system

Correct Answer: Applying operations like stemming and stop word removal to
tokens

Question #443: Which of the following scenarios best illustrates the need for
sophisticated language-specific tokenization?

• A document containing both English and French text.

• A document containing German compound nouns like

"Donaudampfschifffahrtsgesellschaftskapitän".

• A document where capitalization is inconsistent.

• A document with hyphenated words like "state-of-the-art".

• A document with numerical dates in different formats.

Correct Answer: A document containing German compound nouns like

"Donaudampfschifffahrtsgesellschaftskapitän".

Question #444: What is the main difference between Boolean retrieval and ranked
retrieval?

172 | P a g e
 • Boolean retrieval returns an unordered set of documents that match the
query, while ranked retrieval returns an ordered list of documents based on
relevance.

• Boolean is used in small collection and ranked is used in web search

• Boolean retrieval can only handle AND queries while ranked can handle OR

• Boolean retrieval is faster than ranked retrieval.

• Boolean retrieval uses stemming while ranked retrieval doesn't.

Correct Answer: Boolean retrieval returns an unordered set of documents that

match the query, while ranked retrieval returns an ordered list of documents based
on relevance.

Question #445: Level 1: Which data structure is commonly used for efficient
retrieval in Information Retrieval systems?

• Binary Tree

• Linked List

• Inverted Index

• Hash Table

• Array

Correct Answer: Inverted Index

Question #446: Question: What is the main difference between stemming and
lemmatization?

• Stemming uses a vocabulary and morphological analysis, while

lemmatization uses crude heuristic process that chops off the ends of words

• There is no real difference between them

• Stemming is more accurate than lemmatization

• Stemming is faster and more heuristic than lemmatization

173 | P a g e
 • Lemmatization is more commonly used in information retrieval

Correct Answer: Stemming is faster and more heuristic than lemmatization

Question #447: What is the "document frequency" of a term?

• The number of documents in the collection that contain the term

• The average number of times the term appears across all documents

• The length of the postings list for the term

• The total number of times the term appears in the collection

• The number of times the term appears in a specific document

Correct Answer: The number of documents in the collection that contain the term

Question #448: Does using an incidence matrix improve query speed compared to
grepping through text?

• Yes, but only for very small collections of documents.

• No, because modern computers can handle linear scans as fast as any
indexed search.

• Yes, because indexing allows faster lookups without scanning each

document.

• No, because building the incidence matrix requires significant preprocessing

time.

• No, because the incidence matrix is less flexible than grepping for complex
queries.

Correct Answer: Yes, because indexing allows faster lookups without scanning
each document.

Question #449: Which of the following best describes the "Bag of Words" model in
information retrieval?

174 | P a g e
 • A model that represents a document as an unordered collection of words,
disregarding grammar and word order.

• A model that considers the order of words in a document.

• A model that uses a thesaurus to expand the query with synonyms.

• A model that assigns equal weight to all words in a document.

• A model that only considers the first 100 words of a document.

Correct Answer: A model that represents a document as an unordered collection

of words, disregarding grammar and word order.

Question #450: In the context of term weighting, what is the purpose of logarithmic
weighting?

• To ensure that all terms have equal weights.

• To normalize the length of documents.

• To increase the weight of rare terms.

• To reduce the impact of very high term frequencies.

• To eliminate stop words from the index.

Correct Answer: To reduce the impact of very high term frequencies.

Question #451: In the context of spelling correction, what does "isolated-term

correction" mean?

• Correcting multiple terms in a query simultaneously

• Correcting spelling errors based on the surrounding context

• Using phonetic similarity to correct spelling errors

• Correcting a single query term at a time

• Using only the edit distance for correction.

Correct Answer: Correcting a single query term at a time

175 | P a g e
Question #452: When determining the appropriate size of document units for an
information retrieval system, what is a key consequence of the precision vs. recall
trade-off?

• There are no issues with spurious or irrelevant matches.

• Index granularity becomes a significant concern.

• Users will only find relevant documents without any false matches.

• It does not affect retrieval performance.

• The system may miss important passages while increasing the likelihood of
spurious matches.

Correct Answer: The system may miss important passages while increasing the
likelihood of spurious matches.

Question #453: How does the Soundex algorithm handle similar-sounding

consonants?**

• It treats them as vowels

• It maps them to different digits

• It removes them from the term

• It maps them to the same digit based on phonetic equivalence classes

• It ignores them

Correct Answer: It maps them to the same digit based on phonetic equivalence
classes

Question #454: Why is a biword index not the standard solution for indexing terms?

• Using a biword index might seem efficient but can lead to misses due to the
need for more precise patterns.

• A biword index is better than a positional index because it allows for

exhaustive phrase indexing.

176 | P a g e
 • A biword index is sufficient because it effectively handles all relevant term
searches.

• A biword index cannot handle phrases longer than two words, making it less
practical.

• A biword index requires the use of part-of-speech taggers to improve

accuracy but is often used in combination with other methods.

Correct Answer: A biword index is not a standard solution for indexing terms
because it cannot capture all relevant term searches effectively.

Question #455: Which of the following is NOT a typical step in inverted index
construction?

• Index the documents that each term occurs in

• Collect the documents to be indexed

• Perform word segmentation

• Do linguistic preprocessing of tokens

• Tokenize the text

Correct Answer: Perform word segmentation

Question #456: What is a potential drawback of using a biword index for very long
phrases?

• It is less accurate than a positional index

• It cannot handle Boolean queries

• It cannot handle proximity queries

• It can lead to a very large vocabulary size

• It is computationally expensive to build

Correct Answer: It can lead to a very large vocabulary size

177 | P a g e
Question #457: In a system using both a standard inverted index and a permuterm
index, what is the primary role of the permuterm index in processing wildcard
queries?**

• To store the postings lists for all terms.

• To perform stemming and other normalization tasks.

• To calculate the edit distance between the query and vocabulary terms.

• To directly retrieve documents matching the wildcard query.

• To identify the vocabulary terms that potentially match the wildcard pattern,
before using the standard index.

Correct Answer: To identify the vocabulary terms that potentially match the
wildcard pattern, before using the standard index.

Question #458: Why is raw term frequency (tf) not ideal for directly computing
query-document match scores?

• It only considers the presence or absence of a term, not its frequency.

• It is difficult to calculate for large documents.

• It gives too much weight to frequently occurring terms, disproportionately

increasing the score of documents with many occurrences of the term.

• It is only applicable to Boolean retrieval models.

• It doesn't account for the length of the document.

Correct Answer: It gives too much weight to frequently occurring terms,

disproportionately increasing the score of documents with many occurrences of
the term.

Question #459: Question: Which of the following is a characteristic of ranked

retrieval models?

178 | P a g e
 • Results are limited to a small set of documents.

• Users must use a specific query language.

• The system returns an ordering over the documents in the collection with
respect to a query.

• Documents are either a match or not.

• Queries are Boolean expressions.

Correct Answer: The system returns an ordering over the documents in the
collection with respect to a query.

Question #460: You are building a search engine for a large collection of research
papers. Many papers use similar terminology, but the relevance of a paper often
depends on a specific combination of terms. Which weighting scheme would likely
provide the BEST results for ranking documents by relevance?

• Boolean weighting (term present or absent).

• Inverse document frequency alone.

• TF-IDF weighting with cosine normalization, incorporating term proximity.

• Raw term frequency.

• A weighting scheme that ignores document length.

Correct Answer: TF-IDF weighting with cosine normalization, incorporating term

proximity.

Question #461: What does edit distance describe in the context of spelling
correction?

• A matrix used for ranking documents based on term frequency.

• The process of adjusting search results based on contextual information

rather than exact matches.

179 | P a g e
 • The minimum number of insertions, deletions, or replacements needed to
transform one string into another.

• A specialized distance measure that considers the likelihood of character

substitutions.

• The practice of correcting a query by replacing individual characters with

more likely alternatives.

Correct Answer: The minimum number of insertions, deletions, or replacements

needed to transform one string into another.

Question #462: How can asymmetric expansion of query terms be usedfully model
users’ expectations?

• Indexing unnormalized tokens and maintaining a query expansion list are

both methods that require more processing at query time compared to
equivalence classing.

• Normalization can be used to remove diacritics and make tokens consistent,

which can help in matching similar terms regardless of case.

• Query expansion lists can include multiple vocabulary entries to consider for
a certain query term, effectively creating a disjunction of several postings
lists.

• The best amount of equivalence classing or query expansion is a fairly open

Question. Doing some seems a good idea but can easily have unintended
consequences.

• With modern storage costs, the increased flexibility from distinct postings
lists is appealing because it allows for asymmetry in query expansion.

Correct Answer: Query expansion lists can include multiple vocabulary entries to
consider for a certain query term, effectively creating a disjunction of several
postings lists.

Question #463: How does the Vector Space Model handle term weighting?
180 | P a g e
 • Terms are weighted based on their frequency in the document and/or the
collection (e.g., TF-IDF)

• It doesn't; all terms are treated equally

• Terms are weighted based on their grammatical role

• Terms are weighted based on their frequency in the query

• Terms are weighted based on their position in the document

Correct Answer: Terms are weighted based on their frequency in the document
and/or the collection (e.g., TF-IDF)

Question #464: Which of the following is NOT a typical step in pre-processing text
for Information Retrieval?

• Stemming

• Translation

• Stop word removal

• Tokenization

• Normalization

Correct Answer: Translation

Question #465: What is the main purpose of the "Map" phase in the MapReduce
framework for distributed indexing?

• To sort the term-document pairs

• To distribute the document collection into splits and generate (term, docID)
pairs

• To merge postings lists from different machines

• To write the final index to disk

• To assign tasks to worker machines

181 | P a g e
Correct Answer: To distribute the document collection into splits and generate
(term, docID) pairs

Question #466: Question: What is a "stop word" in the context of Information

Retrieval?

• A word that has been stemmed or lemmatized

• A word that is crucial for understanding the meaning of a document

• A word that is very frequent and considered to have little value in selecting
documents

• A word that appears only once in a document collection

• A word that is used to connect other words in a phrase

Correct Answer: A word that is very frequent and considered to have little value in
selecting documents

Question #467: What is the purpose of query optimization in Boolean retrieval?

• To expand the query with synonyms

• To make the query look nicer

• To find the most relevant documents

• To reduce the size of the postings lists

• To determine the best order for processing query terms

Correct Answer: To determine the best order for processing query terms

Question #468: Why might you want to index different parts of an email message
(e.g., body, attachments) as separate documents?

• To allow users to search specifically within those parts, increasing precision

• To improve the accuracy of stemming

182 | P a g e
 • To make it easier to identify the email's language

• To reduce the overall index size

• To automatically classify the email as spam or not spam

Correct Answer: To allow users to search specifically within those parts, increasing
precision.

Question #469: Level 1: What is "Information Need" in the context of Information

Retrieval?

• The user's underlying topic of interest

• The relevant documents retrieved by the system

• The query typed by the user

• The set of all documents in the collection

• The algorithm used for ranking

Correct Answer: The user's underlying topic of interest

Question #470: Level 4: Why is it important to consider document length when

designing a retrieval model?

• Longer documents are always more relevant.

• Document length is irrelevant to retrieval.

• Shorter documents are easier to process.

• Document length is used to calculate precision.

• Longer documents tend to have higher term frequencies, which can skew
results.

Correct Answer: Longer documents tend to have higher term frequencies, which
can skew results.

183 | P a g e
Question #471: Which of the following statements describes a correct technique
for handling general wildcard queries like "se*mon"?

• Create a special inverted index specifically designed for wildcard queries.

• It's impossible to handle general wildcard queries without creating a new

type of inverted index.

• Another approach not mentioned in the context is used to handle general

wildcard queries.

• Use a term-document incidence matrix to combine prefix and suffix

searches.

• To find all terms starting with "se" and ending with "mon", an efficient
approach is to use a B-tree for the prefix and a reverse B-tree for the suffix,
then intersect the results.

Correct Answer: To find all terms starting with "se" and ending with "mon", an
efficient approach is to use a B-tree for the prefix and a reverse B-tree for the suffix,
then intersect the results.

Question #472: What is the main advantage of using cosine similarity over
Euclidean distance in the vector space model?

• Cosine similarity is not affected by document length.

• Cosine similarity provides a distance measure.

• Cosine similarity is more sensitive to differences in term frequency.

• Cosine similarity is faster to compute.

• Cosine similarity can handle negative term weights.

Correct Answer: Cosine similarity is not affected by document length.

Question #473: Difficulty: level1 Question: What is a "stop list" in the context of
Information Retrieval?

184 | P a g e
 • A list of documents that should not be indexed.

• A list of all the tokens in a document.

• A list of misspelled words.

• A list of common words excluded from the index.

• A list of terms used for query expansion.

Correct Answer: A list of common words excluded from the index.

Question #474: You're building a search engine for a language with a very large
character set and frequent spelling variations due to different romanization
schemes (like Chinese). Which combination of techniques would likely be MOST
robust for handling both spelling errors AND wildcard queries in this scenario?

• Soundex and a standard inverted index.

• Edit distance with a B-tree.

• Permuterm index and edit distance.

• A combination of a modified Soundex algorithm (adapted for the language's

phonetics), a k-gram index for wildcard support, and edit distance
calculations (using a language-specific weighting scheme) for refined
spelling correction.

• k-gram index with post-filtering and a permuterm index.

Correct Answer: A combination of a modified Soundex algorithm (adapted for the

language's phonetics), a k-gram index for wildcard support, and edit distance
calculations (using a language-specific weighting scheme) for refined spelling
correction.

Question #475: Why is document frequency (DF) important in some term weighting
schemes (like TF-IDF)?

• DF is not used in term weighting

185 | P a g e
 • High DF indicates a term is likely to be a stop word

• Low DF indicates a term is more discriminative

• High DF indicates a term is more discriminative

• DF is only used in Boolean retrieval

Correct Answer: Low DF indicates a term is more discriminative

Question #476: Level 4: What is an advantage of the Probabilistic Retrieval Model

over the Boolean Model?

• It's simpler to implement.

• It doesn't require an inverted index.

• It's more efficient for very large collections.

• It guarantees retrieval of all relevant documents.

• It provides a ranked list of results based on probability of relevance.

Correct Answer: It provides a ranked list of results based on probability of

relevance.

Question #477: When dealing with text processing for languages like French or
German, which action would improve search accuracy by better handling hyphens
and compound terms?

• Use a compound-splitter module to break down compound nouns into

separate words.

• Perform word segmentation on East Asian languages written without spaces.

• Ignore hyphens and treat them as single words.

• Use apostrophes only for English contractions.

• The system automatically generalizes all forms (e.g., "over-eager" to "over

eager" or "overeager").

186 | P a g e
Correct Answer: The system automatically generalizes all forms (e.g., "over-eager"
to "over eager" or "overeager").

Question #478: How can we modify the k-gram index approach to better handle
wildcard queries like "red*"?

• Expand the wildcard in the k-gram index by treating it as a new k-gram.

• Combine multiple 3-grams from the same word, such as "ing" and "ion" to
match "ingion".

• Issue a prefix-based query in the k-gram index instead of combining two 3-

grams.

• Modify the inverted index to expand wildcards during retrieval.

• Use a range query on the term lengths to find terms of specific lengths
starting with "red".

Correct Answer: Issue a prefix-based query in the k-gram index instead of

combining two 3-grams.

Question #479: Consider a scenario where users often search for product names
that are frequently misspelled, and these misspellings are often not simple
typographical errors (e.g., "Nikkon cammera" for "Nikon camera"). Furthermore,
users sometimes use wildcard queries to find variations of these product names.
Which retrieval model enhancement would be LEAST directly helpful in this
situation?

• Using a query expansion technique based on synonyms and related terms.

• Incorporating edit distance calculations with a low threshold for suggesting

corrections.

• Implementing a phonetic matching algorithm like Soundex.

• Using a k-gram index to handle wildcard queries.

187 | P a g e
 • Building a custom spelling correction module trained on common
misspellings of product names.

Correct Answer: Using a query expansion technique based on synonyms and

related terms.

Question #480: Question: In the context of Boolean query processing, what is the
standard heuristic for processing terms in a conjunctive query (AND of terms)?

• In order of increasing document frequency

• In alphabetical order

• In random order

• Based on the length of the terms

• In order of decreasing document frequency

Correct Answer: In order of increasing document frequency

Question #481: Question: What is the main advantage of using an inverted index for
information retrieval?

• It automatically translates queries into different languages.

• It reduces the storage space required for documents.

• It provides a graphical representation of the document collection.

• It allows for faster query processing.

• It improves the ranking of search results.

Correct Answer: It allows for faster query processing.

Question #482: Question: What is the main goal of stemming and lemmatization?

• To automatically correct spelling errors in queries

• To expand the vocabulary of terms

188 | P a g e
 • To reduce words to their root form

• To translate documents into different languages

• To improve the visual presentation of search results

Correct Answer: To reduce words to their root form

Question #483: What is a potential disadvantage of stemming?

• It only works for English

• It can conflate words with different meanings (over-stemming)

• It slows down query processing

• It increases the size of the index

• It makes it harder to find exact matches

Correct Answer: It can conflate words with different meanings (over-stemming)

Question #484: Which data structure is most suitable for handling wildcard queries
with '*' at the beginning of a term (e.g., *tion)?

• A standard B-tree

• A compressed postings list

• A suffix tree

• A reverse B-tree

• A hashtable

Correct Answer: A reverse B-tree

Question #485: Why might a combination of biword indexes and positional indexes
(as mentioned on slide 16) be a more practical approach for a large-scale search
engine than using only a positional index?

• Biword indexes are easier to update.

189 | P a g e
 • Biword indexes are always more accurate than positional indexes.

• A combined approach can balance query speed and index size.

• Positional indexes are too slow for large datasets.

• Positional indexes cannot handle phrase queries.

Correct Answer: A combined approach can balance query speed and index size.

Question #486: Why is token normalization beneficial for information retrieval

systems?

• Reduces the need for manually creating synonym lists.

• Increases query processing time by allowing for implicit synonym matching.

• Makes it easier to handle variations in tokenization without manual

intervention.

• Reduces storage requirements for term lists.

• All of the above.

Correct Answer: All of the above.

Question #487: Which of the following queries would benefit MOST from using a
biword index?

• "operating system"

• "Vannevar Bush"

• "flights to London"

• "to be or not to be"

• "President of the United States"

Correct Answer: "operating system"

190 | P a g e
Question #488: Level 4: In the context of the Vector Space Model, how is document
similarity typically measured?

• By the Euclidean distance between document vectors

• By the cosine similarity between document vectors

• By the length of the documents

• By the Jaccard coefficient

• By the number of shared terms

Correct Answer: By the cosine similarity between document vectors

Question #489: What is a potential disadvantage of stemming?

• It increases the size of the index

• It can conflate words that should be kept separate, leading to incorrect

matches

• It can decrease recall

• It makes phrase queries impossible

• It can increase precision

Correct Answer: It can conflate words that should be kept separate, leading to
incorrect matches

Question #490: What is a characteristic of the Boolean retrieval model that can be
a disadvantage?

• It is difficult to implement

• It is computationally expensive

• It cannot handle phrase queries

• It often returns too few results

191 | P a g e
 • It often returns too many results, and it's difficult for users to control the
result set size

Correct Answer: It often returns too many results, and it's difficult for users to
control the result set size

Question #491: Which of the following is a disadvantage of using accuracy as an

evaluation measure in information retrieval?

• It only considers relevant documents.

• It is not sensitive to the ranking of documents.

• It is not suitable for skewed data, where most documents are non-relevant.

• It requires relevance judgments.

• It is difficult to calculate.

Correct Answer: It is not suitable for skewed data, where most documents are non-
relevant.

Question #492: Question: What is the main purpose of language identification in

information retrieval?

• To determine the language of documents so that language-specific

processing can be applied

• To determine the sentiment of documents (positive, negative, neutral)

• To translate documents into different languages

• To identify the topic of documents

• To correct spelling errors in documents

Correct Answer: To determine the language of documents so that language-

specific processing can be applied

192 | P a g e
Question #493: In Japanese, how are equivalent forms of the word "Schütze"
represented when written in different scripts and syllabaries?

• Hiragana: シュート、Katakana: シューテ、Chinese Characters: 宿猪

• Hiragana: すゐんが、Katakana: シューテ、Romaji: Sute

• Hiragana: しょうて、Katakana: シュテ、Chinese Characters: 豚豚

• Hiragana: しうんが、Katakana: シューと、Romaji: Sūto

• Hiragana: したず、Katakana: シュート、Romaji: Shu-to

Correct Answer: Hiragana: シュート、Katakana: シューテ、Chinese Characters:

宿猪

Question #494: Why might a user prefer the extended Boolean search capabilities
of a system like WestLaw, despite the availability of ranked retrieval models?**

• Extended Boolean search requires less user expertise.

• Extended Boolean search provides more precise control over the query and
what is retrieved.

• Extended Boolean search is always faster.

• Ranked retrieval models are not suitable for legal documents.

• Ranked retrieval always returns too many results.

Correct Answer: Extended Boolean search provides more precise control over the
query and what is retrieved.

Question #495: Which data structure provides an advantage for efficiently storing
and retrieving a relatively static vocabulary, especially when minor query variants
are not a concern?

• Both hashing and search trees can be used effectively in such cases

• Vector Space Models

193 | P a g e
 • Hashing

• Probabilistic Retrieval

• Search Trees

Correct Answer: Hashing

Question #496: In modern IR systems, why do many systems avoid using stop lists?

• Because stop lists significantly increase storage costs for common words.

• Because modern systems can efficiently handle common words through

compression and term weighting techniques.

• Because stop lists are only used in academic research, not in commercial
applications.

• Because stop lists are inherently incompatible with all modern search
algorithms.

• Because stop lists are required by standard IR system design guidelines.

Correct Answer: Because modern systems can efficiently handle common words
through compression and term weighting techniques.

Question #497: What are the two main components of a basic inverted index?

• Precision and recall

• Dictionary and postings lists

• Documents and queries

• Terms and relevance scores

• Corpus and collection

Correct Answer: Dictionary and postings lists

194 | P a g e
Question #498: Question: Which of the following presents a challenge for
tokenization?

• Converting all text to lowercase

• Handling of punctuation and contractions

• Stemming words to their root form

• Removing stop words from the document

• Determining the language of the document

Correct Answer: Handling of punctuation and contractions

Question #499: How does assigning different weights to edit operations improve
isolated-term spelling correction?

• It ensures that the correct term is suggested based on typical usage

patterns.

• It merely replaces a character without considering its impact on word

meaning.

• It accounts for common keyboard layouts when correcting typos.

• It allows for more accurate misspelling detection.

• It enhances search performance by focusing on substitutions likely to be

intended.

Correct Answer: It enhances search performance by focusing on substitutions

likely to be intended.

Question #500: Which scale of information retrieval systems involves centrally

stored documents and dedicated search machines for collections such as
corporate internal documents or research articles?

• Web Search

• Enterprise/Institutional Search

195 | P a g e
 • Decentralized Search

• Personal Information Retrieval

• Distributed Search

Correct Answer: Enterprise/Institutional Search

Question #501

Question: In the Probabilistic Retrieval Model, documents are ranked based on:

Options:

1. Their publication date

2. The estimated probability that they are relevant to the query

3. Their similarity to the query vector

4. Their length

5. The number of query terms they contain

Correct Answer: The estimated probability that they are relevant to the query

Question #502

Question: Referring to the exercise on slide 14, what is the recommended query
processing order?

Options:

1. (i) AND (ii) AND (iii)

2. (iii) AND (i) AND (ii)

3. (i) AND (iii) AND (ii)

4. Any order for the OR operations, then merge (i) AND (iii), then merge the
result with (ii).

5. (ii) AND (iii) AND (i)

Correct Answer: Any order for the OR operations, then merge (i) AND (iii), then
merge the result with (ii).

196 | P a g e
Question #503

Question: What is meant by "decoding" a document in IR preprocessing?

Options:

1. Creating postings lists

2. Converting a byte sequence into a character sequence, handling character

encodings and document formats

3. Finding synonyms

4. Removing stop words

5. Performing stemming

Correct Answer: Converting a byte sequence into a character sequence, handling

character encodings and document formats

Question #504

Question: Which model is more commonly used in modern web search engines?

Options:

1. Positional Retrieval

2. Ranked Retrieval

3. Proximity Retrieval

4. Extended Boolean Retrieval

5. Boolean Retrieval

Correct Answer: Ranked Retrieval

Question #505

Question: Which of the following is NOT associated with probabilistic retrieval

models in the context of information retrieval?

197 | P a g e
Options:

1. Probabilistic Retrieval

2. Vector Space Model

3. Boolean Retrieval

4. Term-Document Incidence Matrix

5. Inverted Index

Correct Answer: Probabilistic Retrieval

Question #506

Question: What is the typical time complexity of intersecting two postings lists of
lengths x and y using the merge algorithm?

Options:

1. O(x^2 + y^2)

2. O(log(x + y))

3. O(x + y)

4. O(min(x, y))

5. O(x * y)

Correct Answer: O(x + y)

Question #507

Question: Level 4: What is the purpose of using proximity operators (like "NEAR" in
WestLaw) in a query?

Options:

1. To find documents where terms are close to each other within a specified
distance

2. To specify the exact order of terms

198 | P a g e
 3. To exclude documents containing certain terms

4. To broaden the search to include synonyms

5. To increase the weight of specific terms

Correct Answer: To find documents where terms are close to each other within a
specified distance

Question #508

Question: Which of the following is NOT a typical component of an inverted index?

Options:

1. Postings lists

2. Document collection

3. Term frequencies

4. Dictionary of terms

5. Document ID mappings

Correct Answer: Document collection

Question #509

Question: Which of the following best describes the challenge of word

segmentation in East Asian languages when determining the vocabulary of terms?

Options:

1. Word segmentation ensures consistent and unique tokenization every time.

2. Retrieval systems rely on perfect word segmentation for accurate indexing.

3. Character-based indexing completely eliminates the need for word

segmentation.

4. The lack of spaces between words simplifies the indexing process.

199 | P a g e
 5. Multiple possible segmentations can lead to inconsistent or incorrect
tokenization.

Correct Answer: Multiple possible segmentations can lead to inconsistent or

incorrect tokenization.

Question #510

Question: When normalizing tokens in information retrieval, what is the primary

method of grouping different term variations into an equivalence class?

Options:

1. Indexing without normalization.

2. Keeping term forms unchanged.

3. Mapping token variations to a common normalized form.

4. Using character removal rules only.

5. Creating synonym lists for manual grouping.

Correct Answer: Mapping token variations to a common normalized form.

Question #511

Question: What happens when a Boolean query is used without any additional
mechanism to score document matches?

Options:

1. All matching documents are returned in an arbitrary order

2. Documents are ranked based on publication date

3. Only exact matches are returned

4. No documents are returned

5. Documents are ordered by relevance

Correct Answer: All matching documents are returned in an arbitrary order

200 | P a g e
Question #512

Question: What is a "type" in the context of tokenization?

Options:

1. A sequence of characters in a document

2. A stop word

3. A normalized term included in the IR system's dictionary

4. A class of all tokens containing the same character sequence

5. A punctuation mark

Correct Answer: A class of all tokens containing the same character sequence

Question #513

Question: What is the primary reason for determining a suitable document unit for
indexing?

Options:

1. To handle different text directions in languages like Hebrew or Arabic

2. To allow for the combination of multiple files into one document

3. To increase the complexity of search queries

4. To ensure that matches are more relevant and easier to find

5. To simplify the process of creating indices

Correct Answer: To ensure that matches are more relevant and easier to find

Question #514

Question: Which of the following best describes a structure used in information

retrieval that aligns with the dynamic programming approach in edit distance
computation?

201 | P a g e
Options:

1. Retrieval Models

2. Boolean Retrieval

3. Inverted Index

4. Probabilistic Retrieval

5. Vector Space Model

Correct Answer: Inverted Index

Question #515

Question: Which of the following steps is NOT part of the process for converting a
byte sequence into a character sequence?

Options:

1. Converting Unicode characters to binary

2. Ignoring markup in XML files

3. Handling document formats like DOC or ZIP files

4. Decoding the byte sequence using the identified encoding

5. Determining the correct encoding scheme

Correct Answer: Converting Unicode characters to binary

Question #516

Question: Which strategy is recommended for placing skip pointers in a postings

list to optimize intersections?

Options:

1. Not using any skip pointers for efficient comparisons

2. Placing a single skip pointer at each level

3. Placing skips based on random distribution

202 | P a g e
 4. Using skips every P/4 nodes, where P is the length of the postings list

5. Placing skips after every two nodes regardless of P

Correct Answer: Using skips every P/4 nodes, where P is the length of the postings
list

Question #517

Question: Which of the following best describes the unique features mentioned in
the context regarding Japanese writing systems and a specific monetary value?

Options:

1. The challenge of equivalence classing in French due to gender and vowel-

initial distinctions in articles like le, la, l’, and les.

2. The use of hiragana for inflectional endings and function words, with the first
four characters of the final line representing ¥500.

3. The integration of Chinese characters, hiragana, and katakana in modern

Japanese, with the last line expressing a monetary amount equivalent to
¥500,000.

4. The use of katakana for emphasis and the representation of ¥500,000

(¥500,000) through four Chinese characters at the end of the text.

5. The dominance of English on the WWW, with approximately 60% of web

pages written in English.

Correct Answer: The use of katakana for emphasis and the representation of
¥500,000 (¥500,000) through four Chinese characters at the end of the text.

Question #518

Question: In the context of information retrieval, what type of index is most

commonly employed for handling phrase queries when a biword index is not used?

Options:

203 | P a g e
 1. An inverted index

2. A document frequency index

3. A biword index

4. A positional index

5. A term frequency index

Correct Answer: A positional index

Question #519

Question: Which of the following statements best explains why using a positional
index affects the asymptotic complexity of Boolean queries?

Options:

1. It reduces storage requirements by compressing position values.

2. The complexity shifts from being bounded by the number of documents (N)
to the total tokens in the collection (T).

3. Positional indexing is not necessary for supporting phrase searches.

4. Positional indexing increases the number of tokens processed per query.

5. It improves search speed by allowing more efficient retrieval of relevant

documents.

Correct Answer: The complexity shifts from being bounded by the number of
documents (N) to the total tokens in the collection (T).

Question #520

Question: Why is it generally beneficial to process terms in a conjunctive Boolean

query in order of increasing document frequency?

Options:

1. It ensures all relevant documents are retrieved

204 | P a g e
 2. It simplifies the query structure

3. It improves the precision of the query

4. It minimizes the size of intermediate results

5. It maximizes the size of intermediate results

Correct Answer: It minimizes the size of intermediate results

Question #521

Question: Which of the following is used in the Soundex algorithm when retrieving
names based on their Soundex codes?

Options:

1. Boolean Retrieval

2. Inverted Index

3. Probabilistic Retrieval

4. Vector Space Model

5. Term-Document Incidence Matrix

Correct Answer: Term-Document Incidence Matrix

Question #522

Question: Which of the following is a typical example of a stop word?

Options:

1. Index

2. Retrieval

3. Information

4. Algorithm

5. The

205 | P a g e
Correct Answer: The

Question #523

Question: What is an advantage of lemmatization over stemming?

Options:

1. It requires less memory

2. It is more accurate in reducing words to their base form, considering context

3. It is simpler to implement

4. It always improves retrieval performance

5. It is faster

Correct Answer: It is more accurate in reducing words to their base form,

considering context

Question #524

Question: When processing text to create tokens, what should be done with
hyphenated words like "San Francisco-Los Angeles" to ensure accurate retrieval
results?

Options:

1. Use case-sensitive methods to handle hyphens and ensure proper phrase

treatment.

2. Split tokens based on both whitespace and hyphens while maintaining case
insensitivity for phrases.

3. Ignore hyphens and treat them as a single continuous string of characters.

4. Split on hyphens but concatenate the resulting tokens back together,

treating them as one phrase.

5. Split them into two separate words and treat each as a distinct token.

206 | P a g e
Correct Answer: Split tokens based on both whitespace and hyphens while
maintaining case insensitivity for phrases.

Question #525

Question: When considering how users might write queries for words like "Ferrari"
(the brand) versus "ferrari," what is the most likely outcome when using case-
folding?

Options:

1. They will not be considered the same.

2. Users may unintentionally match both terms.

3. It depends on the specific software used.

4. Ferrari and ferrari are treated as equivalent due to case-folding.

5. Case-folding will treat them as different words.

Correct Answer: Ferrari and ferrari are treated as equivalent due to case-folding.

207 | P a g e