Unit-4

Unit-3
 Syllabus Unit-4

• Knowledge based recommendation

• Knowledge representation and reasoning
• Constraint based recommenders
• Case based recommenders
 Knowledge based recommendation

• Knowledge-based recommendation systems (KBRs) provide recommendations by

leveraging domain knowledge, which includes explicit information about users,
items, and their relationships.
• Unlike collaborative filtering, which relies on user interaction history, or
content-based filtering, which focuses on item features,
knowledge-based systems utilize well-defined rules, constraints, and cases
to match user needs with appropriate items.
• This type of system is highly effective in areas where historical data is sparse or
user preferences are complex, such as real estate, healthcare, or high-
involvement purchases (e.g., cars, financial products).
 Knowledge Representation and
Reasoning
• In knowledge-based systems, knowledge representation refers to the method
used to organize and structure domain-specific information (e.g., product
features, user preferences).
• This structured knowledge is then used in reasoning, where the system applies
logical operations or algorithms to infer recommendations based on user input.
 Techniques for Knowledge
Representation:
• Ontology-Based Representation: Ontologies are formal models that describe
relationships between concepts in a domain.
• For example, in a movie recommendation system, an ontology might model
relationships between different genres (e.g., action, drama, comedy), sub-genres,
actors, directors, and movie themes. This allows the system to understand how
different concepts are related, which helps it provide relevant recommendations
based on user preferences.
 Techniques for Knowledge
Representation:
• Rule-Based Systems: Rule-based systems use predefined rules to represent
relationships between users and items. For example:
• "If a user is interested in a car with high fuel efficiency, recommend models
with over 30 miles per gallon.“
• "If a user has a budget between $20,000 and $30,000, show cars priced
within that range." These rules can be manually defined by domain experts or
learned from data through machine learning models.
 Techniques for Knowledge
Representation:
• Decision Trees: Decision trees model user preferences in a hierarchical manner.
• For example, a tree could help a healthcare system recommend treatments by
evaluating symptoms, patient history, and conditions to suggest appropriate
options.
 Reasoning Methods

• Rule-Based Reasoning: In rule-based reasoning, a system evaluates predefined

rules to find items that satisfy user preferences.
• For example, in a financial product recommender, the system might apply rules
like "If a user has a high risk tolerance, recommend aggressive investment
options."
• This approach works well in domains with well-established knowledge and
decision-making processes.
 Reasoning Methods

• Graph-Based Reasoning: Graphs can represent items, users, and their

relationships.
• In a travel recommendation system, for instance, cities, attractions, and hotels
could be modeled as nodes in a graph, with edges representing relationships
(e.g., proximity, user preferences).
• A graph traversal algorithm might recommend places to visit based on proximity
and user interests.
 Example

• In a medical recommendation system, a patient could input symptoms like

"fever, cough, and fatigue."
• The system, using medical knowledge and reasoning, could infer that these
symptoms are common to several conditions, including flu and COVID-19.
• It may apply rules such as "If the patient has a history of respiratory illness,
prioritize flu tests" and recommend specific diagnostic procedures or treatments.
 Constraint-Based Recommenders

• Constraint-based recommenders operate by matching user preferences with

items based on specific constraints.
• These constraints act as filters to eliminate options that don't meet the user's
requirements.
• Constraint-based recommenders are especially useful when users have specific,
inflexible requirements, such as budget limits or product features.
 How Constraint-Based Recommenders
Work
• User Requirements: Users provide explicit input, usually in the form of
requirements or preferences. For example, in a real estate recommendation
system, a user might specify:
• Location: “San Francisco”
• Budget: “$1,000,000 - $1,500,000”
• Number of bedrooms: “3+”
• Type of property: “House”
 How Constraint-Based Recommenders
Work
• Constraint Satisfaction: The system uses these inputs to filter available
properties, recommending only those that meet all the requirements.
• If no property satisfies every constraint, the system might suggest alternatives
(e.g., properties just outside San Francisco or slightly above the budget).
 Techniques for Constraint Satisfaction

• Hard Constraints: These are strict requirements that must be met. For example, if
a user specifies a maximum budget of $1,500,000, the system won’t recommend
any property priced higher.
 Techniques for Constraint Satisfaction

• Hard Constraints: These are strict requirements that must be met. For example, if
a user specifies a maximum budget of $1,500,000, the system won’t recommend
any property priced higher.
• Soft Constraints: These are preferences that can be relaxed if necessary. For
instance, a user might prefer a swimming pool, but if no homes with pools are
available within their budget, the system might recommend homes without
pools.
 Techniques for Constraint Satisfaction

• Hard Constraints: These are strict requirements that must be met. For example, if
a user specifies a maximum budget of $1,500,000, the system won’t recommend
any property priced higher.
• Soft Constraints: These are preferences that can be relaxed if necessary. For
instance, a user might prefer a swimming pool, but if no homes with pools are
available within their budget, the system might recommend homes without
pools.
• Constraint Relaxation: In cases where no items meet all constraints, the system
might suggest relaxing one or more constraints. For example, it might
recommend homes slightly over budget if they meet all other requirements.
 Example: Real Estate Recommender

A user searching for a home in San Francisco might specify the following constraints:
• Budget: “Under $1,500,000”
• Location: “Within 5 miles of downtown”
• Features: “At least 3 bedrooms, modern kitchen, backyard”

The system will then search for homes matching these criteria and might present options like:
• House 1: $1,450,000, 3 bedrooms, 2 miles from downtown, modern kitchen, backyard
• House 2: $1,475,000, 4 bedrooms, 4 miles from downtown, modern kitchen, backyard

If no homes meet all constraints, the system could relax the budget or the distance constraint to
show homes slightly over budget or located 6 miles from downtown
 Case-Based Recommenders

• Case-based recommenders use past examples (or "cases") to make new

recommendations.
• When a user inputs their preferences, the system retrieves similar past cases and
adapts their solutions to the current problem.
• Case-based reasoning is particularly useful when user preferences are complex or
unique, making it hard to rely solely on statistical patterns.
 How Case-Based Recommenders Work

• Case Retrieval: When a user specifies their preferences, the system searches its database
for past cases that are similar to the user’s situation. Each case represents a problem and
its solution. For example, in a car recommendation system, a case might represent a past
user’s purchase decision:
• User 1 bought a compact SUV with a budget of $30,000, prioritizing fuel efficiency
and safety features.
 How Case-Based Recommenders Work

• Case Retrieval: When a user specifies their preferences, the system searches its database
for past cases that are similar to the user’s situation. Each case represents a problem and
its solution. For example, in a car recommendation system, a case might represent a past
user’s purchase decision:
• User 1 bought a compact SUV with a budget of $30,000, prioritizing fuel efficiency
and safety features.
• Case Adaptation: After retrieving similar cases, the system adapts the solution to fit the
new user’s needs. For example, if the new user has a budget of $35,000, the system
might recommend a slightly more expensive SUV with similar features.
 How Case-Based Recommenders Work

• Case Retrieval: When a user specifies their preferences, the system searches its database
for past cases that are similar to the user’s situation. Each case represents a problem and
its solution. For example, in a car recommendation system, a case might represent a past
user’s purchase decision:
• User 1 bought a compact SUV with a budget of $30,000, prioritizing fuel efficiency
and safety features.
• Case Adaptation: After retrieving similar cases, the system adapts the solution to fit the
new user’s needs. For example, if the new user has a budget of $35,000, the system
might recommend a slightly more expensive SUV with similar features.
• Case Learning: The system continuously adds new cases to its database, improving its
ability to make future recommendations.
 Example: Car Recommender

• Imagine a user looking for a car with a budget of $40,000, prioritizing safety and
comfort.
• The system retrieves a case from a previous user with a similar budget and
preferences, who purchased a 2021 Honda Accord.
• It then adapts the recommendation based on the current user’s needs,
suggesting a 2022 model or a similar sedan that prioritizes safety and comfort,
like the Toyota Camry.
 Applications

• Job Matching: A case-based recommender might match a user’s resume to

previous job seekers with similar qualifications and suggest jobs that worked for
them.
• Customer Support: Case-based reasoning is frequently used in customer support
systems, where the system retrieves solutions to past problems similar to the
user’s current issue.