Recommendation Algorithm Based Onprobability Matrix Factorization

In recent years, In order to solve the problem of information overload, recommender systems have developed rapidly. As a branch of collaborative filtering recommendation, probabilistic matrix factorization, is widely studied by the academic circle in recent years. With the increasing in the number of users and projects, the evaluation data is becoming sparser. Probability matrix factorization technique is mainly used in the user program rating matrix data, but it doesn’t take the relationships of trust and focus between users into consideration, which leads to the decrease of recommended performance. In this paper, we will take the rating matrix and relationships of users’ trust as data source and making further study in the probability matrix factorization technique.First of all, aiming at the big error problem about calculating Pearson between users in the sparse data, considering of all the users’ score, the common score between users and the average score for goods, we bring up a new method to calculate similarity between users asymmetric similarity calculation method, which is a kind of relative similarity.Secondly, between the trusted user information, the probability matrix factorization technique can score matrix factorization projects, at the same time, considering the asymmetric similarity between users’ information and properly introducing the social information to probability matrix factorization model so that potential feature vector of users in the process of factorization of matrix can be affected by the similar potential feature vector of similar friends, which can make it closer to the ideal potential feature vector of users.Thirdly, we propose a propulsion model for enhance the trust relationship between users, which is based on Social MF model, taking the information of asymmetry similar trust between the trusted users, using the asymmetric similarity to judge whether there is a common preference between users who have trust relationship and putting the preference information into the existing social network, to enhance the trust network.Finally, we give the experimental evaluations and analysis of the algorithms proposed in this paper, compare the performance between the proposed algorithms and other existing algorithms.