Regularizing Matrix Factorization with User and Item Embeddings for Recommendation