We modify the model of Bayesian Probabilistic Matrix Factorization. We fuse social relations and item contents with rating data in a novel way. The proposed method gets more accurate results in faster converge speed. The proposed method can alleviate data sparsity problem and cold‐start problem.
