Dual-regularization Matrix Factorization Recommendation System Based On Review Neural Modeling And Social Network Constraints

Recommender system provides personalized services and support for users by obtaining users’ personal needs,interest preferences and social relations.However,limited by data sparsity and cold start,the traditional matrix factorization method has been difficult to meet the needs of application development.As information sources such as reviews,ratings,and social networks become more abundant,how to mine and utilize this information to improve the performance of the recommendation system has become an urgent problem to be solved.In this paper,based on solving the problem of users’ rating prediction of items,a new recommendation system is proposed by using the auxiliary information such as comment text and social network.The main work of this paper is as follows:A dual-regularized matrix factorization(Dual-regularized Matrix Factorization based on hierarchical attention,DMF)model based on deep modeling of review text is proposed.First embed the review text of the user or item,and then send it to the Bi GRU layer to enhance the relevance of the words before and after the long text;then,use a twolayer attention mechanism to assign the weight of the contribution of different reviews to the intermediate representation vector.The PMF graph model is used to fuse user and item representation vectors,complete the unknown scoring matrix,and use alternate joint optimization and back propagation methods to complete the training of model parameters.In order to effectively use social network information,improve the accuracy of scoring prediction.On the basis of DMF,assuming that the embedded representation of users is affected by their trusted users,they join social network constraints to form the So DMF model.In particular,in order to measure the trust between users,the User-User,User-Item,and User-User-Item graphs are decomposed and constructed,and the random walk algorithm with restart is used to solve them.Through the experimental analysis of four data sets,it is confirmed that DMF improves the user comment modeling effect through the double-layer attention mechanism,thereby significantly reducing the scoring error.On this basis,So DMF relies on social network constraints to further improve the scoring accuracy.