Abstract: In recent years, the personalized context-aware recommendation is the rub and hotness in the research of recommendation system, and the data sparseness is the main problem faced by the current recommendation systems. In the setting of music recommendation, the representing method of varieties of situational information is improved. A model of TFPMF is proposed, which combines the model of RR-PMF with the tensor decomposition. TFPMF is optimized by alternative least squares (ALS). By the simulation experiments in the last.fm dataset, we got the TOP-N recommended list through the simulation program. The simulation results show that the proposed algorithm has great advantages in the evaluation index of Precision, Recall and NDCG, and the algorithm can effectively alleviate the data sparsity problem.

Key words: recommendation system, reciprocal rank, probabilistic matrix factorization, tensor factorization, TFPMF (Tensor Factorization-based Probabilistic Matrix Factorization)