Chapter 11. Finding hidden genres with matrix factorization

 

The matrix is only numbers, and this chapter is about the matrix and how to create one:

  • You’ll learn about dimensionality reduction recommender algorithms.
  • Reducing similarity will help you find latent (hidden) factors in the data.
  • You’ll train and use a singular value decomposition (SVD) to create recommendations.
  • You’ll learn how to fold in new users and items into an SVD.
  • You’ll look at another matrix factorization model called the Funk SVD, which is more flexible than the original SVD.

What have you learned so far? In chapter 8, we looked at collaborative filtering using neighbor-based filtering. In this chapter, we’re going to return to collaborative filtering, but this time we’re not talking about neighborhoods. Instead, we’ll explore latent factors. In chapter 10, we talked about latent factors, but at that point, we talked about latent factors in the content data. Now we’ll look at latent factors in relation to collaborative filtering, which means in behavioral data.

11.1. Sometimes it’s good to reduce the amount of data

11.2. Example of what you want to solve

11.3. A whiff of linear algebra

11.3.1. Matrix

11.3.2. What’s factorization?

11.4. Constructing the factorization using SVD

Diagonal matrix

Reducing the matrix

How much should the matrix be reduced?

Predict a rating

Solving the problem of the zeros in the rating matrix using imputation

11.4.1. Adding a new user by folding in

11.4.2. How to do recommendations with SVD

11.4.3. Baseline predictors

11.4.4. Temporal dynamic