Lucky Talk: Low-regret Recommendations and Item-item Collaborative Filtering
summary
Matrix of binary likes, random user per round, bandit feedback and regret defined as average 1 - like; initial recommendations behave like random guessing with slope about 2−2μ until a cold-start ≈ O(d) period. Algorithm samples many users in parallel to estimate column inner products, builds an epsilon-net clustering of similar items (item-item collaborative filtering) and then lets each user refine which clusters they like. This shared parallel similarity test yields sublinear regret after the cold-start phase.
AI-written summary · about
chapters (4)
related talks
58mRecommender Systems – A Walkthrough
37mA primer on recommendation systems.
33mPeople who liked this talk also liked ... Building Recommendation...
41mRecommender Systems: Crafting Effective Personalised Experiences
32mLet's talk about Recommendation Systems!
33mPicking What to Watch Next: Build a Recommendation System
29mRecommender Systems for Personalized Financial Advice
20mBuilding Recommender system