Collaborative filtering, matrix factorization, two-tower retrieval and the cold start problem.
8h · 7 lessons · 0 challenges
Before this: Unsupervised Learning
Recommendation is the highest-revenue application of ML and a standing favorite in system design interviews. This module covers the classical stack — collaborative filtering and matrix factorization — then the modern two-tower retrieval and ranking split that industrial systems use.
Cold start, feedback loops and popularity bias get real treatment, since those are the follow-up questions.
A new user arrives with no history and you have six interactions to get them to stay. Strategy against a simulated audience.
Worth 450 XP.
