Weighted Slope One Algorithm Based On Probabilistic Matrix Factorization

With the rapid development of the information age,the amount of information that people are exposed to is growing at an exponential level.In order to make users find the information they need as quickly as possible,the recommendation system emerges.This paper conducts in-depth research on the Slope One recommendation algorithm.This algorithm is a memorybased collaborative filtering algorithm that is widely used in recommendation system because of its high recommendation efficiency,easy to understand and easy to implement.However,the original Slope One algorithm does not take into account the user similarity and the item similarity,so the recommendation quality needs to be improved.To solve this problem,this paper incorporates the user similarity and the item similarity as the weight in the calculation of the Slope One algorithm,which aims to improve the accuracy of the Slope One algorithm’s rating prediction.For the similarity calculation,instead of using sparse scoring matrix,this paper uses the user feature matrix and the item feature matrix obtained by Probabilistic Matrix Factorization algorithm(PMF).Since these two matrices can dig out more implicit information of users and items,they are more suitable for the similarity calculation.Therefore,this paper combines the Probabilistic Matrix Factorization algorithm(PMF)and the Slope One algorithm,proposes a Weighted Slope One algorithm based on Probabilistic Matrix Factorization(PMF).In the experiment,this paper explores the optimal value of algorithm parameters,and then compares the improved algorithm with the Slope One algorithm(SO),the Weighted Slope One algorithm(WSO)and other common recommendation algorithms.The results show that the three algorithms proposed in this paper have significantly improved the accuracy of predicting user’s rating of unknow items.