1 · Recommender foundations
What a recommender does, the user-item rating matrix, explicit vs implicit feedback, and the two problems that define the field: sparsity and cold-start.
●Track 13 · live
Recommender Systems
How Netflix, Spotify & Amazon decide what's next — CF, matrix factorization, ranking.
The engine behind every 'recommended for you' feed: content-based filtering, memory-based collaborative filtering (user-user / item-item), matrix factorization (SVD, ALS/SGD, implicit feedback), ranking evaluation (precision@k, MAP, NDCG), and the beyond-accuracy goals (diversity, coverage, cold-start). Every algorithm is built from scratch in NumPy / scikit-learn and runs in your browser via Pyodide.
17 / 17 lessons6 chapters
2 · Content-based filtering
Recommend by item features: build item profiles, weight them with TF-IDF, and match them to a user's taste with cosine similarity.
3 · Collaborative filtering (memory-based)
'Users like you also liked…': user-user and item-item neighborhoods, the similarity metrics that power them, and rating prediction.
4 · Matrix factorization
The breakthrough that won the Netflix Prize: latent factors, truncated SVD, and learning factors with ALS / SGD.
5 · Evaluation & ranking
Recommenders are ranking problems: leak-free splits, top-k metrics (precision@k, recall@k, MAP, NDCG), and the beyond-accuracy goals.
6 · Modern recommenders & capstone
Implicit feedback and BPR ranking loss, hybrid + context-aware recommenders, and a full end-to-end recommender capstone.