The latent factor and matrix factorization models are limited to capture non-linear patterns of the user and latent spaces. However, VAE[1] solve this problem by finding the latent factor distribution after projecting User or Item Ratings, and using the latent factor distribution to generate the corresponding rating predictions.