What is matrix factorization in recommendations (SVD, ALS)?

Matrix factorization is a method in recommender systems that allows representing interactions between users and items (such as movies, products) as a product of two or more matrices with smaller dimensions.

• SVD (Singular Value Decomposition): decomposes the user-item matrix into three matrices, highlighting hidden factors that explain user preferences and item characteristics. This helps predict ratings for pairs where data is missing.

• ALS (Alternating Least Squares): an optimization algorithm for matrix factorization that alternately updates user and item matrices, minimizing prediction error. It is often used for sparse data.

Example: if we have a matrix of user ratings for movies, matrix factorization can reveal hidden preferences (genres, styles) and recommend movies that the user hasn't seen yet but might like.