Recommendation systems have evolved from collaborative filtering methods that identify similar users or items based on past interactions, to latent factor models using matrix factorization that represent items and users as vectors in a factor space to overcome sparsity and similarity limitations.