📄️ Recommender systems
Premium course on recommender systems: neighborhoods, matrix factorization, two-tower models, cold start, NDCG and position bias. 6 hours, 40-question exam.
📄️ 1. Framing the problem
Module 1 of the Recommender Systems premium course: predicting a rating vs ranking a list, explicit and implicit feedback, sparse matrices, and the real business objective.
📄️ 2. Neighborhood CF
Module 2 of the Recommender Systems premium course: similarities, neighborhoods, weighted prediction, user vs item, compute cost, extreme sparsity and shrinkage.
📄️ 3. Matrix factorization
Module 3 of the Recommender Systems premium course: latent factors, SGD and ALS, regularization, user and item biases, implementation on the catalog with implicit or Surprise.
📄️ 4. Content-based filtering
Module 4 of the Recommender Systems premium course: description embeddings, user profile, cosine similarity, the filter bubble, and when content beats collaborative filtering.
📄️ 5. Hybrid approaches
Module 5 of the Recommender Systems premium course: weighting, switching, enriching factors with content, and choosing which approach takes over when.
📄️ 6. Two-tower model
Module 6 of the Recommender Systems premium course: the two-tower model, negative sampling, retrieval then ranking, and approximate nearest-neighbor search.
📄️ 7. Cold start
Module 7 of the Recommender Systems premium course: content for new items, questionnaire and popularity for new users, controlled exploration and cold-start metrics.
📄️ 8. Ranking metrics
Module 8 of the Recommender Systems premium course: precision and recall at k, NDCG computed by hand, catalog coverage, diversity and novelty, and why RMSE misleads.
📄️ 9. Implicit feedback
Module 9 of the Recommender Systems premium course: clicks as a noisy signal, confidence weighting, position and popularity bias, and propensity-based correction.
📄️ 10. Full project
Module 10 of the Recommender Systems premium course: end-to-end pipeline on the catalog, temporal split, metrics table, online A/B plan and known limits.
📄️ Recap and exam
Complete recap of the Recommender Systems premium course: framing, CF, factorization, content, hybrids, two-tower, cold start, ranking metrics, implicit bias, then the 40-question exam.