A Collaborative Filtering Recommendation Algorithm Based on Probabilistic Matrix Factorization

Abstract

The information of users and items is difficult to obtain due to expensive expert labeling costs and privacy issues. So the sequential behavior relationship of consumption is introduced to solve the problem of collaborative filtering recommendation. The sequential behavior relationship is extended to the item consumption network, and the calculation method of the asymmetric item-item similarity is defined. The item-item matrix is constructed via probabilistic matrix factorization to explore the potential neighborhood information of items. The neighbor information is integrated into the user-item rating matrix, and the user-item rating matrix is reorganized by matrix factorization to predict the user ratings for items. A collaborative filtering recommendation framework model based on user and item neighbors is proposed. Experiments on real data sets show that the collaborative filtering method based on two-level probability matrix factorization integrating sequential behavior can improve the accuracy of the user's rating prediction and the performance of recommendation.