Collaborative Filtering Algorithm Research Based on Matrix Factorization and Muti-path Trust Degree Fusion
Hanmin Ye, Qiuling Zhang, Peiliang Huang · 2016
According to the recommendation quality is not high and cold start problem of the recommendation system in the case of sparse data, a collaborative filtering algorithm based on the combination of matrix decomposition technique and social network trust model is proposed. First of all, in the degree of trust computing, expert node method is introduced to determine the existence of multiple paths of trust degree between two non adjacent nodes. At the same time, in order to improve the item rating matrix prediction accuracy of users, under the base of the matrix decomposition optimization, the regularization method is introduced. Then, the multi path trust degree matrix is fused with the user item scores obtained by the matrix regularization to make the score prediction. Finally, the proposed algorithm is validated and compared with the RMSE value on the MovieLens two data sets of different sizes, and the results show that the recommendation accuracy of the proposed algorithm is obviously superior to the traditional algorithm.