Recommender Systems
How platforms decide what to show you — collaborative filtering, matrix factorization, two-tower retrieval, session-based models, diversity, cold start, and ad ranking.
7 lessons
01The Recommendation Problem
Predicting what a user will want — explicit vs implicit feedback, content-based vs collaborative filtering, the cold-start problem, and the ranking metrics (recall@k, NDCG) that judge a recommender.
Matrix Factorization
The model-based collaborative filtering workhorse — learning latent user and item factors, the SVD/ALS connection, regularization, and how it handles the sparse rating matrix that won the Netflix Prize.
Deep & Two-Tower Recommenders
Scaling recommendation to billions of items — neural collaborative filtering, the two-tower retrieval architecture, the retrieval-then-ranking funnel, and serving with approximate nearest neighbors.
Session-Based & Real-Time Recommendations
Moving beyond static user profiles — sequential models that read what you're doing right now, real-time feature pipelines, and contextual bandits that balance exploitation with exploration.
Diversity, Cold Start & Exploration
The three problems every recommender faces beyond ranking: what to show new users/items (cold start), how to avoid filter bubbles (diversity), and how to discover unknown preferences (exploration).
Ad Ranking & CTR Prediction
The ML engine behind digital advertising — predicting click-through rates with deep cross-feature models, the mechanics of the second-price auction, and budget-aware bid optimization.
Recommender Systems Quiz
Check your understanding of feedback types, ranking metrics, matrix factorization, two-tower retrieval, session models, diversity, and ad ranking.