Module 35

Recommender systems

Advanced module (after graduation): rating table, similar users and movies, matrix factorization, honest evaluation

When you finish this module you can say:

I can turn a list of ratings into a user-movie table, build recommendations from similar users, similar movies and matrix factorization, and measure them honestly on held-out ratings with RMSE and hit@5.

By the end of this module you will be able to build your own movie recommender from a list of ratings, compare two methods honestly on held-out ratings, and measure the variety of the recommendations as well as their accuracy.

Lessons

6 lessons · ≈ 61 min

1. The user-movie table9 min
2. Similar users10 min
3. Similar movies10 min
4. Matrix factorization10 min
5. Honest evaluation10 min
6. Project: a movie recommender12 min · project

Practice exercises

Short writing exercises, independent of the lessons: Go to the exercises

1. How full is the table
2. Subtract the user average
3. Error on held-out ratings