UNIT-1

1. Definition of Information Retrieval System (IRS)

An Information Retrieval System (IRS) is a software system designed to store, manage,

retrieve, and organize information from large collections of data based on users’ queries.
Unlike traditional database systems that return exact matches, IRS focuses on finding
relevant information that satisfies imprecise or ambiguous user needs, especially in text or
multimedia documents.

Key Characteristics:

Relevance-based Retrieval

Returns documents or information ranked by how relevant they are to the query.

Handles Unstructured Data

Works mainly with text, images, videos, and other non-tabular data types.

Supports Query Languages

Allows users to express natural language or Boolean queries.

Indexing and Ranking

Uses indexing techniques to quickly retrieve data, and ranking algorithms to sort results by
relevance.

Example Systems:

 Web search engines like Google, Bing

 Library catalog systems
 Digital archives
 Enterprise document search systems

2. Objectives of Information Retrieval Systems

The main goals of an IRS are:

The main objective of an Information Retrieval System is to help users find relevant
information quickly and accurately from a large collection of unstructured or semi-structured
data. Below is a detailed explanation of the core objectives of an IRS:

1. Retrieve Relevant Information

 The primary goal is to find and deliver documents or records that are most relevant to the
user's query.
 IRS does not always look for exact matches but focuses on semantic relevance—
meaning how closely the content of a document matches the intent of the query.
Example:

If a user searches for "machine learning applications", the IRS should fetch documents
related to real-world uses of machine learning, even if the exact words aren't used.

2. Efficient Storage and Organization of Data

 The IRS must store a large amount of information efficiently, usually in the form of
documents, web pages, or multimedia.
 It should provide structured ways (like indexing) to organize this data for quick access
and retrieval.

Example:

 Search engines maintain indexes of billions of web pages to retrieve results in

milliseconds.

3. Fast and Accurate Retrieval

 Speed is a critical factor—users expect search results instantly.

 IRS uses inverted indexes, caching, and ranking algorithms to ensure that searches are
both fast and accurate.

Objective: Achieve a balance between speed and quality of results.

4. Ranking of Results Based on Relevance

 Among many possible results, the IRS must rank the most relevant ones at the top.
 This is typically based on:
 Frequency of keywords (TF-IDF)
 Position of terms
 Link analysis (e.g., PageRank)
 User behavior (click-through rates)

Goal: Minimize user effort in finding the best result.

5. Support for Various Query Types

 IRS should allow users to search using:

 Keywords
 Natural language
 Boolean logic
 Fuzzy queries
 It should also suggest corrections, auto-complete, or expand queries for better
results.

Example: A user types "artifical inteligence", and the system suggests "Did you mean:
artificial intelligence?"

6. Adaptability and Learning from User Feedback

 Modern IRS aim to learn from user interactions.

 Feedback like clicks, likes, dwell time, and skipped results help the system improve
relevance over time using machine learning.

7. Multilingual and Multimedia Support

 IRS should support multilingual documents and multimedia types like:

 Images (e.g., Google Images)

 Videos (e.g., YouTube Search)
 Audio (e.g., Spotify search)

Goal: Enable retrieval from diverse formats, not just plain text.

8. Access Control and Security

 Especially in enterprise and government applications, the IRS should ensure:

 Authorization: Only authorized users can access sensitive data.
 Privacy: Protect users’ search behavior and personal data.

9. Scalability and Maintainability

 The IRS should be able to handle:

 Growth in data volume
 Increased number of users
 Real-time updates and deletions

3. Functional Overview of an IRS

The functional overview of an Information Retrieval System (IRS) refers to the core
processes and components that work together to collect, store, process, retrieve, and rank
information in response to a user’s query.

Below is a detailed explanation of the key functional components and their roles:

1. Document Acquisition / Data Collection

 This is the first step where data (documents, web pages, PDFs, multimedia, etc.) is
gathered from various sources.
 Sources may include:
 Web crawling (for search engines)
 Digital libraries
 Databases or local storage
 User submissions

 Function: Gather raw data that needs to be made searchable.

2. Document Processing / Preprocessing

Before data can be stored or searched, it needs to be processed and cleaned.

 Common preprocessing steps:
 Tokenization: Breaking text into words, terms, or tokens.
 Stop Word Removal: Removing common words like "and", "the", "is" that carry
little meaning.
 Stemming/Lemmatization: Reducing words to their base form (e.g., "running" →
"run").
 Normalization: Converting all text to lowercase, removing punctuation or
formatting.
 Metadata Extraction: Extracting title, author, date, etc.

 Function: Prepare raw data for indexing and retrieval.

3. Indexing

 Indexing is a core functionality that allows fast searching.

 An inverted index is commonly used, which maps each term to a list of documents where
it appears.

Example:

Term Document IDs

"machine" 1, 3, 5

"learning" 1, 2, 5

 Function: Create a searchable structure to improve retrieval speed and accuracy.

4. Query Processing

 When a user inputs a query, it also needs to be processed similarly to documents.

 Query processing includes:

 Tokenization
 Stop word removal
 Query expansion (adding synonyms)
 Spell checking and suggestion
 Parsing Boolean operators (AND, OR, NOT)

 Function: Transform user input into a form that can be matched against the index.

5. Matching / Retrieval

 The system compares the processed query with the index to find documents that match
the query terms.

 Matching techniques:
 Boolean Matching: Matches based on AND, OR, NOT
 Vector Space Model: Measures similarity between query and document vectors
 Probabilistic Models: Calculates the probability of relevance
 Function: Retrieve a set of documents that are potentially relevant.

6. Ranking and Scoring

 Once matching documents are found, they are ranked by relevance using scoring
algorithms.

 Ranking factors may include:

 TF-IDF (Term Frequency – Inverse Document Frequency)

 PageRank
 User behavior (clicks, time spent)
 Recency of the document

 Function: Present the most relevant results at the top of the results list.

7. User Interface (UI)

 The UI allows users to:

 Enter queries
 View and refine search results
 Apply filters (date, author, category)
 Provide feedback (likes, saves)

 Function: Ensure a user-friendly experience for searching and interacting with the
system.

8. Feedback and Learning (Optional in Modern IRS)

 Modern IRS use user feedback to improve search results over time.

 Feedback types:
 Explicit (user ratings)
 Implicit (clicks, time spent)

 Function: Learn and adapt to user behavior using machine learning algorithms.

9. Access Control and Security (For Enterprise/Confidential Systems)

 IRS may restrict access to documents based on:

 User role
 Confidentiality level
 Login authentication

 Function: Maintain privacy and security in document retrieval.

Need to update

4. Relationship to Database Management Systems (DBMS)

Feature IRS DBMS

Data Type Unstructured/Semi-structured Structured (tables)
Query Language Natural language, Boolean, vector space SQL
Matching Approximate, relevance-based Exact match
Examples Search engines, library catalogs Banking, HR, logistics systems

Conclusion:
While both systems store and retrieve data, IRS is optimized for information discovery, and
DBMS is optimized for data accuracy and integrity.

5. Digital Libraries and Data Warehouses

Digital Libraries

 A collection of digitized content including books, articles, images, and multimedia.

 Uses IRS techniques to allow search, browse, and retrieval.
 Examples: JSTOR, IEEE Xplore, Google Books.

Data Warehouses

 Centralized repositories of integrated data from multiple sources.

 Primarily used for analytical processing (OLAP).
 Employs structured querying and reporting, not typical IRS capabilities.

IRS in Digital Libraries:

 Supports fuzzy matching, multilingual search, metadata-based filtering.

IRS vs Data Warehousing:

  IRS is user-centric and retrieval-oriented.
 Warehouses are analytics-centric and structured.

Information Retrieval System Capabilities

1. Search Capabilities

a. Keyword Search

 Match user-specified keywords with indexed documents.

b. Boolean Search

 Use operators: AND, OR, NOT.

 Example: “machine AND learning NOT deep”

c. Phrase Search

 Enclose in quotes: “information retrieval”

d. Proximity Search

 Find terms within a specified distance.

 Example: “data NEAR/3 mining”

e. Wildcard and Truncation

 Use * or ? for partial matches.

 Example: “comput*” matches “computer”, “computing”.

f. Fuzzy Search

 Matches approximate spellings (e.g., Levenshtein distance).

2. Browse Capabilities

a. Hierarchical Browsing

 Navigate through categories or taxonomies (e.g., subject > sub-topic > article).

b. Faceted Browsing

 Filter content using metadata facets like author, year, topic.

c. Hyperlink Navigation

 Follow embedded links in documents (e.g., hyperlinks in Wikipedia).

d. Thesaurus-Based Browsing
  Explore related concepts using a controlled vocabulary.

3. Miscellaneous Capabilities

a. Relevance Feedback

 User marks results as relevant/irrelevant; system refines search.

b. Query Expansion

 System adds synonyms or related terms to broaden search.

c. Multilingual Support

 Cross-language retrieval and translation tools.

d. Personalization

 Tailor results based on user profile or history.

e. Document Clustering and Summarization

 Group similar documents and generate abstracts.

f. Ranking Algorithms

 Use TF-IDF, BM25, PageRank, or neural networks to rank results.

 UNIT-2

Cataloguing and Indexing

1. History and Objectives of Indexing

History of Indexing:

 Ancient libraries (e.g., Library of Alexandria): Manual indexing using scroll labels
and catalogues.
 Renaissance & Print Era: Alphabetical indexes in books for navigation.
 Modern Era: Library classification systems (e.g., Dewey Decimal, Library of
Congress).
 Digital Era: Computational indexing in search engines and digital libraries.

Objectives of Indexing:

 To identify and represent content for efficient retrieval.

 To organize documents in a retrievable structure.
 To reduce search space by mapping documents to relevant keywords or concepts.
 To enhance relevance in information retrieval through effective query matching.

2. Indexing Process

Indexing transforms documents into a searchable representation.

Steps in Indexing:

1. Document Collection: Input documents from text, images, etc.

2. Tokenization: Breaking text into meaningful units (tokens).
3. Stop-word Removal: Eliminate common words (e.g., “the”, “is”).
4. Stemming/Lemmatization: Reducing words to root forms (e.g., “running” → “run”).
5. Term Weighting: Assigning importance (e.g., TF-IDF scores).
6. Index Creation: Storing term-document associations for retrieval.

Types of Indexing:

 Manual Indexing: Human-assigned keywords and descriptors.

 Automatic Indexing: Performed by software systems.

3. Automatic Indexing

Automatic Indexing is the process of assigning index terms without human intervention.

Advantages:

 Fast and scalable.

 Consistent across documents.
  Adaptable to multiple domains.

Techniques:

 Statistical Methods: Term frequency (TF), inverse document frequency (IDF).

 Natural Language Processing (NLP): POS tagging, entity recognition.
 Machine Learning Approaches: Supervised classification, clustering.

4. Information Extraction (IE)

IE is the process of automatically extracting structured information from unstructured data.

Key Tasks in IE:

 Named Entity Recognition (NER): Identify names, places, organizations.

 Relationship Extraction: Find relations (e.g., “X founded Y”).
 Event Extraction: Identify actions and events.

Applications:

 Building knowledge graphs.

 Summarization.
 Semantic search.

Data Structures in Information Retrieval

1. Introduction to Data Structures

In IRS, efficient data structures are used for storing and accessing large volumes of textual
data.

Key Roles:

 Fast search and retrieval.

 Compact representation of data.
 Support for query processing.

2. Stemming Algorithms

Stemming reduces words to their base form.

Popular Algorithms:

 Porter Stemmer: Most widely used rule-based algorithm.

 Lovins Stemmer: Early stemmer with a large suffix list.
 Snowball Stemmer: Improved and customizable version of Porter.

Example:

Words like “connected”, “connecting”, “connection” → “connect”

3. Inverted File Structure

An inverted file (or inverted index) is a fundamental data structure in IRS.

Structure:

 Vocabulary (dictionary): All unique terms.

 Postings list: For each term, list of document IDs where it appears.

Example:
Term Postings List

data Doc1, Doc2, Doc5

mining Doc1, Doc3

Advantages:

 Quick full-text search.

 Efficient Boolean and ranked retrieval.

4. N-Gram Data Structures

N-Grams are contiguous sequences of n items from a text.

Types:

 Character N-grams: “text” → {“te”, “ex”, “xt”} (bigrams)

 Word N-grams: “data mining techniques” → {“data mining”, “mining techniques”}

Applications:

 Fuzzy search.
 Spell correction.
 Language modeling.

5. PAT Tree (PATRICIA Tree)

A compressed trie used for indexing strings.

Features:

 Efficient prefix matching.

 Space-saving over traditional tries.
 Useful in natural language processing tasks.

6. Signature File Structure

A filtering technique for text retrieval.

Process:

 Create a bit-vector signature for each document using hashing.

  Query is also converted into a signature.
 Perform bitwise comparison to filter candidate documents.

Advantages:

 Faster filtering.
 Suitable for large text corpora.

7. Hypertext and XML Data Structures

Hypertext:

 Documents contain links to other documents (e.g., HTML pages).

 Enables non-linear navigation.
 Supports anchor text indexing.

XML:

 Extensible Markup Language used for structured data.

 Requires tree-based parsing and indexing.
 XPath/XQuery used for retrieval.

Indexing XML:

 Tag-based indexing.
 Element-path and value-path indexes.

8. Hidden Markov Models (HMMs)

A statistical model used to describe sequences (used in IE and speech recognition).

Structure:

 States (hidden)
 Observations (visible)
 Transition and emission probabilities

Applications in IR:

 Named entity recognition.

 Part-of-speech tagging.
 Word prediction and disambiguation.
 UNIT-3

Automatic Indexing

1. Introduction

Automatic Indexing refers to the process of using algorithms to assign index terms (or
keywords) to a document without human intervention. The main goal is to improve the
efficiency, scalability, and objectivity of the indexing process.

2. Classes of Automatic Indexing

There are several approaches to automatic indexing, typically grouped as follows:

A. Statistical Indexing

 Based on term frequency and co-occurrence patterns.

 Uses statistical properties of words in documents.
 Examples: TF-IDF, Mutual Information, Chi-square tests.

B. Natural Language Indexing

 Utilizes syntactic and semantic analysis.

 Considers parts of speech, grammar, phrase structure.
 Involves tools like POS taggers, parsers, and lemmatizers.
 Example: Indexing only nouns and noun phrases.

C. Concept Indexing (Semantic Indexing)

 Goes beyond keywords to extract concepts using:

o Thesauri
o Ontologies
o Word sense disambiguation
 Captures synonymy and polysemy.
 Example: "AI", "Artificial Intelligence", and "Machine Intelligence" mapped as the
same concept.

D. Hypertext Linkages

 Leverages links between documents (e.g., HTML links, citations).

 Indexing includes anchor text and neighboring pages.
 Basis for algorithms like PageRank.
 Useful in web IR systems and Wikipedia-style resources.

3. Comparison of Indexing Approaches

Feature Statistical Natural Language Conceptual Hypertext

Accuracy Moderate High High Moderate

Complexity Low High Very High Moderate

 Feature Statistical Natural Language Conceptual Hypertext

Speed Fast Slower Slowest Fast

Context Awareness None Some High Medium

Document and Term Clustering

1. Introduction to Clustering

Clustering is the process of grouping similar documents or terms based on content, structure,
or statistical relationships.

Applications in IR:

 Improve retrieval efficiency and relevance.

 Enable thematic browsing.
 Assist in query expansion.
 Organize search results (cluster-based navigation).

2. Thesaurus Generation

A thesaurus in IR is a collection of terms with semantic relationships.

Automatic Thesaurus Generation:

 Uses statistical co-occurrence of terms across documents.

 Builds semantic networks:
o Synonymy
o Hierarchical relations (broader/narrower terms)
o Associative relations

Techniques:

 Clustering
 Co-word analysis
 Latent Semantic Analysis (LSA)

3. Manual vs. Automatic Clustering

Aspect Manual Clustering Automatic Clustering

Performed by Humans Algorithms

Accuracy High (domain-specific) Moderate to high

Scalability Low High

Cost High Low (after setup)

4. Automatic Term Clustering

Focuses on grouping terms that appear in similar contexts.

Methods:

 Co-occurrence Matrix: Rows and columns are terms; values indicate co-appearance.
 Cosine Similarity: Measures similarity between term vectors.
 Hierarchical Clustering: Terms grouped based on similarity.

5. Complete Term Relation Method

A comprehensive method where all possible relations between terms are examined.

Process:

 Build co-occurrence matrix.

 Compute term-term similarity.
 Create relation graph (nodes = terms, edges = relatedness).
 Cluster based on graph structure.

Drawback: Computationally expensive for large vocabularies.

6. Clustering Using Existing Clusters

Instead of building clusters from scratch, new items are assigned to existing ones.

Approaches:

 Nearest Neighbor Assignment

 Centroid Similarity Matching
 K-Means Incremental Update

7. One-Pass Assignment

Also known as online clustering, it processes each document/term once:

Steps:

1. Assign to the most similar existing cluster.

2. If no suitable cluster found, create a new one.
3. Update cluster centroid as new items arrive.

Advantages:

 Fast and memory-efficient.

 Suitable for streaming data.

8. Item Clustering (Document Clustering)

Focuses on grouping similar documents rather than terms.

Types:

 Flat Clustering: All clusters at the same level (e.g., K-means).

 Hierarchical Clustering: Builds a tree (dendrogram) of nested clusters.

Distance Measures:

 Cosine similarity
 Euclidean distance
 Jaccard index

9. Hierarchy of Clusters

Hierarchical clustering builds multi-level structures of related items.

Types:

 Agglomerative (Bottom-Up): Start with individual items, merge into clusters.

 Divisive (Top-Down): Start with all items in one cluster, split recursively.

Advantages:

 Better representation of relationships.

 Supports browsing from general to specific topics.

Example Dendrogram:
Root

/ \

Sci Arts

/ \ \

AI Bio Music

UNIT-4

Advanced Concepts in Automatic Indexing

1. Search Statements and Binding

Search Statements:

 Formal expressions constructed to retrieve information.

 Often use:
o Keywords
o Boolean Operators: AND, OR, NOT
o Modifiers: Truncation (*), proximity, wildcards

Binding:

 Determines how terms and operators are grouped.

 Operator Precedence: Parentheses control binding.
 Example:
o ((data AND mining) OR (machine AND learning)) NOT healthcare

Binding ensures accurate interpretation of user queries.

2. Similarity Measures and Ranking

Similarity measures compare queries and documents to estimate relevance.

Common Similarity Measures:

Method Description

Cosine Similarity Measures angle between vector representations

Jaccard Coefficient Ratio of common to total terms

Euclidean Distance Measures absolute distance between vectors

TF-IDF Weighting Highlights important and unique terms

BM25 Probabilistic model considering term frequency and document length

Ranking:

 Based on similarity scores

 Documents sorted in descending order of relevance
 Improves user satisfaction

3. Relevance Feedback

Allows the system to improve search results based on user feedback.

Types:

 Explicit Feedback: User marks relevant/irrelevant results.

 Implicit Feedback: System tracks clicks, dwell time.
Techniques:

 Rocchio Algorithm: Adjusts query vector by incorporating relevant and irrelevant

documents.

Updated Query Vector:

Q_new = αQ + βD_rel - γD_irrel

Relevance feedback boosts accuracy and personalization.

4. Selective Dissemination of Information (SDI)

SDI delivers relevant, personalized information updates to users based on their interest
profiles.

Key Concepts:

 User Profile: Stored representation of topics of interest.

 Document Matching: New documents are matched against profiles.
 Notification Mechanism: Email alerts, RSS feeds, dashboards.

Applications:

 Academic literature alerts (e.g., PubMed)

 News aggregators
 Market research platforms

5. Weighted Searches of Boolean Systems

Enhancement of Boolean search using weights to reflect importance of terms.

Weighted Boolean Model:

 Terms and operators have assigned weights (e.g., 0 to 1)

 Combines logical operators with relevance ranking

Benefits:

 More flexible than strict Boolean logic

 Allows partial matches and scoring

Example:
scss
CopyEdit
(data^0.8 AND mining^1.0) OR (statistics^0.5)

6. Searching the Internet and Hypertext

Challenges of Internet Search:

 Scale and heterogeneity of web content

 Lack of structure in HTML pages
 Dynamic and duplicate content

Hypertext Search Concepts:

 Uses hyperlink structure

 Anchor text indexing improves search relevance

Important Algorithms:

 PageRank: Measures importance of a page based on incoming links.

 HITS (Hyperlink-Induced Topic Search): Identifies hubs and authorities.

Modern Techniques:

 Semantic search
 Snippet generation
 Query suggestions/autocompletion

Information Visualization

1. Introduction to Information Visualization

Information Visualization is the graphical representation of abstract data to enhance human

understanding and decision-making.

Goals:

 Reveal patterns, trends, and outliers

 Support exploration of large datasets
 Aid analytical reasoning

Examples:

 Word clouds
 Heatmaps
 Graph visualizations
 Timelines and hierarchical trees

2. Cognition and Perception

Understanding how users perceive and process visual information is critical in visualization
design.
Perceptual Principles:

 Pre-attentive Processing: Immediate perception of basic visual features (color,

shape, orientation)
 Gestalt Principles: Human tendency to group visual elements (proximity, similarity,
continuity)

Cognitive Considerations:

 Working Memory Limits: Humans can hold 5–9 items at once.

 Cognitive Load: Simpler visuals are easier to interpret.
 Contextual Understanding: Prior knowledge affects interpretation.

3. Information Visualization Technologies

A. Tools and Libraries:

Tool/Library Description

Tableau Commercial tool for interactive dashboards

[Link] JavaScript library for creating custom visualizations

Power BI Microsoft’s data visualization suite

Gephi Open-source network graph visualization

Plotly, Matplotlib, Seaborn Python libraries for data plots

B. Visualization Techniques:
Type Description Example

Time Series Line charts, timelines Stock prices over time

Hierarchical Tree maps, dendrograms File systems, taxonomies

Geospatial Maps with overlays Crime mapping, COVID-19 tracking

Network Graphs Nodes and edges Social networks, citation graphs

Multivariate Scatter plots, heatmaps Correlation matrices

C. Interactive Visualization:

 Brushing and linking

 Zoom and filter controls
 Real-time updates

UNIT-5
Text Search Algorithms

1. Introduction to Text Search Techniques

 Definition: Techniques used to locate occurrences of a pattern (query) within a large

body of text.
 Applications: Search engines, text editors, DNA sequencing, plagiarism detection.
 Key challenges: Efficiency, scalability, handling large data volumes, approximate
matching.

2. Software Text Search Algorithms

a) Naïve Search Algorithm

 Checks for pattern occurrence at every position.

 Time complexity: O((n-m+1)*m), where n = text length, m = pattern length.
 Simple but inefficient for large texts.

b) Knuth-Morris-Pratt (KMP) Algorithm

 Preprocesses pattern to create a "longest prefix suffix" (LPS) array.

 Avoids unnecessary re-examination of characters.
 Time complexity: O(n + m).
 Good for exact matching.

c) Boyer-Moore Algorithm

 Searches pattern from right to left.

 Uses bad character and good suffix heuristics to skip sections.
 Usually faster than KMP in practice.
 Best-case complexity sublinear, worst-case O(n + m).

d) Rabin-Karp Algorithm

 Uses hashing to compare pattern hash with text substring hashes.

 Efficient for multiple pattern searches.
 Average complexity: O(n + m), worst case O(nm).

3. Hardware Text Search Systems

 Specialized hardware accelerates text search operations.

 Use in high-throughput systems (network intrusion detection, large-scale databases).
 Examples:
o Content Addressable Memory (CAM): Parallel pattern matching.
o FPGA-based systems: Custom hardware for pattern matching.
o GPU acceleration: Parallelize search algorithms for big data.

Multimedia Information Retrieval

1. Overview

 Multimedia IR deals with retrieval of audio, video, images, and other non-textual
data.
 Challenges:
o Data heterogeneity
o Feature extraction complexity
o Semantic gap between data and user intent

2. Spoken Language Audio Retrieval

 Retrieval based on spoken content.

 Techniques:
o Automatic Speech Recognition (ASR)
o Phonetic indexing and matching
o Keyword spotting
 Applications: Video captioning, voice assistants.

3. Non-Speech Audio Retrieval

 Focus on sounds like music, environmental noises, alarms.

 Feature extraction from pitch, timbre, rhythm.
 Applications: Music search, sound effect libraries.

4. Graph Retrieval

 Retrieval of graph-structured data (social networks, chemical structures).

 Queries match subgraphs or patterns.
 Uses graph matching algorithms.

5. Imagery Retrieval

 Content-based image retrieval (CBIR).

 Features used: color histograms, texture, shape.
 Techniques: feature extraction, similarity matching.

6. Video Retrieval

 Combines audio, visual, and text metadata.

 Key frames extraction, shot segmentation.
 Semantic concept detection.

Information System Evaluation

1. Introduction

 Evaluation assesses system effectiveness in retrieving relevant information.

 Important to improve and benchmark IR systems.

2. Measures Used in System Evaluation

 Measure Description

Precision Fraction of retrieved documents that are relevant

Recall Fraction of relevant documents that are retrieved

F-measure Harmonic mean of precision and recall

Mean Average Precision (MAP) Average precision across queries

ROC Curve Plots true positive rate vs false positive rate

3. Measurement Example: TREC Results

 Text Retrieval Conference (TREC): Standardized benchmarking platform for IR.

 Provides large test collections and relevance judgments.
 Results allow comparison of different IR approaches.
 Example metrics reported: Precision at rank 10 (P@10), MAP.