Research On Recommendation Algorithm Based On Probability Matrix Factorization

The emergence of personalized recommendation system for e-commerce site has brought a better opportunity for development,at the same time,with the rapid development of the network technology,customers and businesses more and more to pursue a higher recommended quality.Therefore,the recommendation algorithm has been extensively studied.In this paper,probabilistic matrix factorization algorithm is studied,and the methods are combined with SVD algorithm and clustering model respectively from the aspects of user and product,the main work is as follows:1.A probabilistic matrix factorization algorithm based on trust relationship is proposed.First,the initial scoring matrix is preprocessed by adding commodity popularity and time-weighted factors.Then,an effective method to calculate the trust relationship between users is presented:Using the new rating matrix and an improved method to obtain the similarity between users,and we also consider the user's influence in social relationships to obtain the trust matrix.At last,the trust matrix between users is added to the objective function of probabilistic matrix factorization algorithm for finding the potential eigenvectors of users and items,and use them to predict the missing rates.A probabilistic matrix factorization algorithm based on trust relationship is established.Experiments with Epinions dataset and Movielens dataset show that the proposed algorithm can effectively improve the accuracy of recommendation.2.A probabilistic matrix factorization algorithm based on inter-commodity relationship is proposed.In this algorithm,the similarities between the products are found out,including the similarities obtained from the scoring matrix,the similarities between the product tags and the similarities between the product categories,and then incorporated into the objective function of the probability matrix factorization algorithm.By adding the objective function,the potential vectors between products are optimized to predict the missing score.Experiments show that,considering the relationship between products can reduce the recommended error,and the improved algorithm can improve the recommendation accuracy.