Matrix Factorization for Recommender Systems

Total time needed: ~4 hours
Learning Objectives
This list will go over three matrix factorization methods (unconstrained, SVD, and NMF) used in Recommender Systems.
Potential Use Cases
Recommending movies (Netflix), "watch next" (YouTube), or "items you might like" (Amazon).
Target Audience
Go through the following annotated items in order:
VIDEO 1. Introduction to Matrix Factorization
  • Overview of the use of matrix factorization applied to Netflix recommendation
35 minutes
OTHER 2. Lecture Notes on Unconstrained Matrix Factorization Methods from Carnegie Mellon
  • What is the theory behind unconstrained matrix factorization?
60 minutes
ARTICLE 3. SVD for Recommender Systems
  • How to implement SVD for recommender systems?
30 minutes
VIDEO 4. Video Lecture on SVD for Recommender Systems from Stanford
  • What is the theory behind SVD for recommender systems?
15 minutes
OTHER 5. Lecture Notes on Non-negative Matrix Factorization from Stanford
  • What is the theory behind non-negative matrix factorization?
60 minutes
LIBRARY 6. Python Package (NIMFA) to solve NMF
  • How do I solve NMF and other factorization problems in Python
30 minutes

Concepts Covered