Statistics for Machine Learning

Matrix factorization using the alternating least squares algorithm for collaborative filtering

Alternating least squares (ALS) is an optimization technique to solve the matrix factorization problem. This technique achieves good performance and has proven relatively easy to implement. These algorithms are members of a broad class of latent-factor models and they try to explain observed interactions between a large number of users and items/movies through a relatively small number of unobserved, underlying reasons/factors. The matrix factorization algorithm treats the user-item data (matrix dimensions m x n) as a sparse matrix and tries to reconstruct with two lower-dimensional dense matrices (X and Y, where X has dimensions m x k and Y has ...