Recommender MidTerm -2

Part A
1. List the basic components of content based systems

Feature extraction: Identifies attributes of items.

Similarity measurement: Compares item features with user

preferences.

2. Recall gini index

The Gini Index is a metric used to measure the impurity or

diversity of a dataset. A lower Gini Index indicates higher purity.

3. Spell about normalized deviation

Normalized deviation measures the deviation of a value from

the mean, scaled by a factor (e.g., range or standard deviation), to make
comparisons across datasets.

4. What are push and nuke attacks

Push attack: Increases the ranking of specific items by giving

them artificially high ratings.

Nuke attack: Lowers the ranking of specific items by giving them

artificially low ratings.

5. Spell about sequential ensembles

Sequential ensembles are models where base learners are built

sequentially, with each model improving on the errors of the previous
one. Example: AdaBoost.

6. Tell about the feature usage in recommender system

User features: Preferences, demographics, or past behavior.

Item features: Attributes like genre, price, or category.

 7. What is metrics?

Metrics are measures used to evaluate the performance or

quality of a system, model, or process.

8. Define social tagging

Social tagging is the process of users adding labels or keywords (tags)

to online content, making it easier to organize and discover.

Part B
1. Illustrate rule based classifers

Definition

● Rule-based classifiers use a set of "if-then" rules for classification.

● Each rule maps input features to a class label.

Working Principle

● Rules are learned from training data or predefined by experts.

● Classification is based on matching input features to these rules.

Advantages

● Easy to interpret and understand.

● Can handle both numerical and categorical data.

Disadvantages

● Limited flexibility in complex scenarios.

● May require many rules for high accuracy.

2. Show the importance of learning user profiles and filtering

Personalized User Experience

● Tailors content based on user preferences.

● Increases user engagement and satisfaction.

Efficient Information Access

● Filters out irrelevant content.

● Helps users find what they need faster.
 Better Decision Making

● Improves recommendations and predictions.

● Supports targeted advertising and marketing strategies.

3.Interpret persistent personalization in knowledge based system

Definition

● Persistent personalization refers to adapting the system to user

needs over time.
● It retains user preferences and behavior across sessions.

How It Works

● The system collects and stores user data for continuous learning.
● Personalization is updated with each interaction, improving
recommendations.

Benefits

● Enhances user experience by providing relevant content.

● Builds a deeper understanding of user needs, leading to better
decisions.

4. Explain about bayes classifier

Definition

● Bayes classifier is a probabilistic classifier based on Bayes'

Theorem.
● It predicts the class of an object by calculating the likelihood of
each class given the features.

Working Principle

● It computes the probability of each class based on input features.

● The class with the highest probability is chosen as the prediction.
 Advantages and Disadvantages

● Simple and efficient, especially for large datasets.

● Assumes feature independence, which might not always hold
true.

5. Outline the method of feature argumentation hybrids

Definition

● Feature augmentation hybrids combine multiple feature sets to

improve model performance.
● The goal is to enhance the data representation by adding relevant
features from different sources.

Working Principle

● Original features are augmented with new or transformed features.

● New features can be extracted using domain knowledge or machine
learning techniques.

Benefits and Challenges

● Improves model accuracy by capturing more information.

● May increase computational complexity and risk of overfitting.

6. Interpret different evaluation paradigms in recommendation systems

Offline Evaluation

● Evaluates the system using historical data, without real-time

interaction.
● Common metrics include accuracy, precision, recall, and F1 score.

Online Evaluation

● Involves real-time user interaction and feedback to evaluate system

performance.
● Metrics like click-through rate (CTR) and user engagement are used to
assess success.
 Hybrid Evaluation

● Combines both offline and online evaluation to get a comprehensive

view.
● Provides a balance between simulation accuracy and real-world
performance.

7. Demonstrate trust walker system

Definition

● Trust Walker is a system for evaluating and managing trust in

online environments.
● It is based on a trust network, where users share trust
relationships with others.

Working Principle

● The system uses a graph-based approach to propagate trust

values through the network.
● Trust is computed based on the relationships and behaviors of
connected users.

Applications and Benefits

● Used in recommendation systems, e-commerce, and social

networks to ensure reliable interactions.
● Enhances system security by filtering out untrustworthy or
malicious entities.

8. Summarize the trade offs in attacks models

Complexity vs. Accuracy

● More complex attack models can simulate realistic threats but

are harder to implement.
● Simpler models may be easier to use but might not capture all
attack strategies.

Speed vs. Coverage

 ● Fast attack models provide quick results but may lack
comprehensive coverage.
● Detailed models take longer to run but give a deeper insight into
vulnerabilities.

Generalization vs. Specificity

● Generalized models can cover a wide range of attacks but may

miss specific vulnerabilities.
● Specialized models are highly accurate for certain attacks but
lack broader applicability.

Part C

1. Explain High level architecture of content-based recommender systems

Content-based recommender systems rely on the characteristics

(content) of items and users to generate personalized recommendations. The
system predicts the user’s interest in an item by comparing the content of
items with the user’s preferences. Below is an explanation of the high-level
architecture and its components.

1. Key Components

a. Input Data

● Item Profiles: A collection of attributes or metadata describing the

items (e.g., genre, director, cast for movies).
● User Profiles: A representation of user preferences, either explicitly
provided (e.g., ratings) or inferred from historical data.

b. Similarity Computation

● Measures how closely the features of an item match the user profile.
● Common Similarity Measures:
○ Cosine Similarity: Measures angle similarity between feature
vectors.
○ Euclidean Distance: Measures straight-line distance in feature
space.
○ Pearson Correlation: Measures linear correlation between
features.
 c. Recommendation Engine

● Core Task: Ranks items based on similarity scores.

● Output: A list of recommended items that match the user’s
preferences.

2. Workflow

1. User Data Collection:

○ Collect explicit feedback (ratings, likes) or implicit feedback
(clicks, watch history).
2. Item Feature Representation:
○ Build profiles for each item using features extracted from
metadata.
3. User Profile Generation:
○ Construct user profiles by analyzing past interactions or
preferences.
4. Similarity Matching:
○ Match the user profile with item profiles using similarity
measures.
5. Ranking and Recommendations:
○ Rank items based on computed similarity scores and
recommend top items to the user.

3. Example: Content-Based Recommender for Movies

● Input Data:
○ Movie profiles: Genre (Action, Comedy), Director (Spielberg), Cast
(Tom Hanks).
○ User profile: Interested in Comedy and movies by Spielberg.
● Feature Extraction:
○ Movie and user preferences converted into feature vectors.
● Similarity Computation:
○ Cosine similarity matches the user profile with the feature vectors
of movies.
● Output:
○ Recommends comedy movies directed by Spielberg, such as
Catch Me If You Can.

2. outline on constraint-based recommender systems

Constraint-based recommender systems suggest items based on

specific user-defined constraints or logical rules, rather than relying solely on
user behavior or product attributes. These systems intelligently refine user
inputs to ensure recommendations align with preferences and constraints.

1. Introduction to Constraint-Based Recommender Systems

● Definition: Suggests items by applying predefined logical rules or

constraints provided by users.
● Example: In real estate, if a user specifies a need for a house with at
least three bedrooms, the system filters out properties that don't meet
this criterion.
● Key Difference from Search Filters:
○ Constraint-based systems use logical rules like "Family size ≥ 5
means 3+ bedrooms."
○ Normal search filters apply direct product attribute matching.

2. How Constraint-Based Systems Work

Step 1: User Input Users provide preferences or constraints (e.g., family size,
ZIP code).

Step 2: Rule Matching The system checks its knowledge base for rules
matching the user's input. Example: If a user specifies a family size of 6,
the system may infer additional constraints like "3+ bedrooms" and "2+
bathrooms."

Step 3: Query Refinement Logical rules expand the user’s input to refine
the search query.

Step 4: Item Retrieval The system retrieves items that satisfy all
constraints. Example: A user specifies a large family and a ZIP code; the
system refines the query to include "3+ bedrooms" and "Price ≥
$100,000."

4. Handling Conflicting Constraints

● When constraints cannot be met simultaneously (e.g., "Max Price =

$100,000" and "Min Bedrooms = 5"), the system suggests repair
proposals to relax constraints.
● Example: Suggest reducing the number of bedrooms or increasing the
maximum price.
 5. Adding More Constraints

● If too many results are retrieved, the system suggests additional

constraints based on historical data or patterns from similar users.
● Example: After specifying price and bedrooms, the system may propose
adding constraints on bathrooms based on common preferences.

3. illustrate the goals of recommender system

Recommender systems are designed to suggest items or content to

users in a manner that aligns with both user satisfaction and business
objectives. While the primary goal is to increase sales and profits for the
merchant, achieving this involves addressing several operational and technical
goals that enhance user experience and engagement.

1. Primary Goal: Relevance

● Definition: The system's primary objective is to recommend items that

are most relevant to the user’s preferences.
● Importance: Users are more likely to interact with and consume items
they find interesting, driving sales and satisfaction.

2. Secondary Goals

a) Novelty

● Definition: Ensuring that the recommended items are new and

previously unseen by the user.
● Purpose: Prevents redundancy and keeps the user engaged with fresh
suggestions.
● Example: Suggesting newly released movies in a user's preferred genre
rather than popular classics they might already know.

b) Serendipity

● Definition: Providing unexpected yet delightful recommendations that

align with latent interests.
● Purpose: Expands user preferences and introduces them to new types
of items.
 c) Increasing Recommendation Diversity

● Definition: Ensuring that the list of recommendations contains varied

items to cater to different preferences.
● Purpose: Reduces the risk of user dissatisfaction by providing options
across diverse categories.

3. Soft Goals

a) Improving User Satisfaction

● Purpose: Satisfied users are more likely to return, increasing loyalty and
site usage.

b) Gaining Insights into User Preferences

● Purpose: Helps merchants understand user needs better, allowing

them to personalize offerings and strategies.

c) Providing Explanations for Recommendations

● Purpose: Enhances transparency and trust by showing users why

certain items were recommended.

4. summarize the methods in evaluating offline recommender systems

Evaluating offline recommender systems involves several metrics and

methodologies to assess their effectiveness, efficiency, and user impact. Below
is a summary of the key evaluation methods:

1. Accuracy

● Accuracy measures how well the recommender system predicts user

preferences.
● Metrics like Mean Absolute Error (MAE) or Root Mean Squared Error
(RMSE) quantify the difference between predicted and actual ratings.
● High accuracy ensures users receive relevant recommendations.
2. Coverage

● Coverage evaluates the proportion of items or users that the system can
recommend to.
● User-space coverage: The fraction of users for whom
recommendations are made.
● Item-space coverage: The fraction of items included in
recommendations.
● Ensures the system addresses data sparsity and provides inclusive
recommendations.

3. Confidence and Trust

● Measures the system's reliability in its predictions.

● Includes reporting confidence intervals for ratings, allowing users to
assess the certainty of recommendations.
● Algorithms with narrower confidence intervals and accurate predictions
improve trustworthiness.

4. Novelty

● Novelty focuses on recommending items that users have not seen

before.
● Promotes discovery of new interests and expands user engagement.
● Typically measured via user feedback or online experimentation.

5. Serendipity

● Serendipity emphasizes surprising but relevant recommendations.

● These recommendations introduce users to new experiences beyond
their usual preferences.
● Example: Suggesting an Ethiopian restaurant to a user familiar with
Indian cuisine.

6. Diversity

● Ensures that the recommended items vary in content or characteristics.

● Reduces redundancy in recommendations and increases the likelihood
of user satisfaction.
● Measured using similarity scores between recommended items in a list.
7. Robustness and Stability

● Evaluates the system's resilience to fake ratings or changing user

behavior.
● Robust systems maintain reliable recommendations even when
attacked or when patterns in data evolve.

8. Scalability

● Assesses the system's ability to handle large datasets efficiently.