Hands-On Machine Learning with C++

Factorization algorithms

It would be nice to describe the interests of the user with more extensive features—not in the format of they love movies X, Y, and Z, but in the format of they love romantic comedies. Besides the fact that it increases the generalizability of the model, it also solves the problem of having a large data dimension—after all, the interests are described not by the items vector, but by a significantly smaller preference vector.

Such approaches are also called spectral decomposition or high-frequency filtering (since we remove the noise and leave the useful signal). There are many different types of matrix decomposition in algebra, and one of the most commonly used is called Singular Value Decomposition (SVD). Initially, ...