Recommendation Systems –
Lecture 1
Dr. Aditya Gupte
What Are Recommendation Systems?
1.A Recommendation System (RS) is an intelligent information filtering system that
predicts user preferences and suggests relevant items.
[Link] helps users discover items (products, movies, news, services) that match their
interests.
[Link] reduces information overload in digital platforms.
[Link] uses historical data, user behavior, item features, or domain knowledge.
[Link] is widely used in e-commerce, streaming platforms, social media, job portals, and
education platforms.
[Link] main objective is personalization at scale.
[Link] improves user engagement, satisfaction, and revenue.
• Example:
• Netflix recommending movies.
• Amazon recommending products.
• Spotify recommending playlists.
Why Are Recommendation Systems
Important?
[Link] internet contains massive volumes of content (millions of items).
[Link] cannot manually search through all available options.
[Link] improves user experience significantly.
[Link] use RS to increase sales through cross-selling and upselling.
[Link] improves customer retention and loyalty.
[Link] enables targeted marketing.
[Link] creates competitive advantage for digital platforms.
• Real Example:
• 80% of Netflix watch time comes from recommendations.
• Amazon generates significant revenue from its recommendation engine.
Where Are Recommendation Systems Used?
1.E-commerce: Product recommendations (Amazon).
[Link]: Movie/music recommendations (Netflix, Spotify).
[Link] Media: Friend suggestions (Facebook).
[Link] Portals: Job matching (LinkedIn, Indeed).
[Link] Learning: Course recommendations (Coursera).
[Link] Platforms: Personalized news feeds (Google News).
[Link]: Personalized treatment suggestions.
Basic Components of a Recommendation
System
[Link]: Individuals interacting with the system.
[Link]: Products, movies, songs, articles, etc.
[Link]-Item Interaction Matrix: Ratings, clicks, purchases.
[Link] Model: Algorithm to estimate user preference.
[Link] Engine: Produces ranked item list.
[Link] Loop: Explicit (ratings) or implicit (clicks).
[Link] Mechanism: Measures performance.
Types of Feedback in Recommendation
Systems
[Link] Feedback: User gives ratings (1–5 stars).
[Link] Feedback: Clicks, watch time, purchase history.
[Link] feedback is accurate but sparse.
[Link] feedback is abundant but noisy.
[Link] systems mostly rely on implicit data.
[Link] systems combine both.
[Link] is used for continuous model improvement.
Taxonomy of Recommendation Systems
[Link] Filtering
[Link]-Based Filtering
[Link]-Based Systems
[Link] Systems
[Link]-Aware Recommenders
[Link]-Based Systems
[Link] Learning-Based Recommenders
What is Collaborative Filtering?
[Link] on the idea: “Users with similar preferences will like similar
items.”
[Link] not require item features.
[Link] user-item interaction data.
[Link] similarity between users or items.
[Link] well when large data is available.
[Link] with cold start problem.
[Link] popular in industry.
• Example:
If Aditya and Rahul both liked Movie A and B,
and Rahul liked Movie C,
→ Recommend Movie C to Aditya.
User-Based Collaborative Filtering
[Link] similarity between users.
[Link] similarity measures (Cosine, Pearson).
[Link] nearest neighbors.
[Link] rating based on neighbor ratings.
[Link] ratings to reduce bias.
[Link] well when user preferences are stable.
[Link] expensive for large systems.
Item-Based Collaborative Filtering
[Link] similarity between items.
[Link] scalable than user-based CF.
[Link]-computes item similarity matrix.
[Link] by Amazon.
[Link] because item similarity changes less.
[Link] well for large user bases.
[Link] to maintain in production.
Limitations of Collaborative Filtering
[Link] Start Problem (new user/item).
[Link] sparsity (most ratings missing).
[Link] issues.
[Link] bias.
[Link] attacks (fake ratings).
[Link]-specialization.
[Link] bubbles.
What is Content-Based Recommendation?
[Link] items similar to what user liked.
[Link] item features (genre, keywords, tags).
[Link] user profile.
[Link] user profile with item features.
[Link] independently of other users.
[Link] user cold-start better.
[Link] diversity.
Steps in Content-Based Filtering
[Link] item features.
[Link] items as feature vectors.
[Link] user profile.
[Link] similarity between user and items.
[Link] items.
[Link] top-N items.
[Link] user profile dynamically.
Example of Content-Based Filtering
User likes:
• Action movies
• Sci-fi genre
• High IMDb rating
System recommends:
• Avengers
• Interstellar
• The Matrix
Because they match feature vector.
What is Knowledge-Based Recommendation?
[Link] domain knowledge.
[Link] not rely on historical ratings.
[Link] for expensive products.
[Link] constraints or rules.
[Link] used in real estate, cars.
[Link] cold start effectively.
[Link] domain expertise.
Types of Knowledge-Based Systems
[Link]-Based Recommenders.
[Link]-Based Recommenders.
[Link]-based filtering.
[Link]-based systems.
[Link] recommenders.
[Link]-tree-based recommenders.
[Link] elicitation systems.
Why Evaluation is Important?
[Link] measure prediction accuracy.
[Link] compare algorithms.
[Link] avoid overfitting.
[Link] assess user satisfaction.
[Link] improve ranking quality.
[Link] measure business impact.
[Link] validate system performance.
Accuracy Metrics
[Link] (Root Mean Square Error)
[Link] (Mean Absolute Error)
[Link]
[Link]
5.F1 Score
[Link]
[Link] Rate
Beyond Accuracy
[Link]
[Link]
[Link]
[Link]
[Link]
[Link]
[Link] Trust