Collaborative filtering, content-based retrieval and the two-stage system they become.
7h · 9 lessons · 0 challenges
Before this: Unsupervised Learning & Anomaly Detection
Recommenders are the most widely deployed machine learning there is, and they are built from a factorization you can derive. This module starts at the rating matrix, derives collaborative filtering from its cost, then handles the two awkward realities: implicit feedback and users with no history.
Then content-based filtering and the two-tower retrieval that scales it, which is the architecture behind most feeds you use.
A new user with three clicks. Recommend the fourth, and explain what your system is betting on.
Worth 350 XP.
