Research On Social Network Recommender Systems Based On Non-negative Multiple Matrix Factorization

With the growing number of Internet users worldwide,Internet information has shown explosive growth.Recommender system can help users identify their most interesting resources in the mass of information,and it is an effective software tool to solve the problem of information overload.A lot of different recommendation methods have been put forward in academia,and have been widely used in industry.Because recommender systems have a wealth of practical application and great commercial value,the research topic is well received in computer science.With the development of Web2.0,the study of Recommender Systems Based on social networks has gradually begun.The social information between users can effectively improve the recommendation quality of recommender systems and provide users with more accurate personalized recommendation services.This paper first introduces the current research situation of recommender systems at home and abroad,and summarizes the existing theoretical results and existing problems of current social network recommendation systems.Then,the modeling and analysis of the social network are carried out,and the non-negative multi matrix factorization algorithm and the user link prediction method are discussed in detail.According to the characteristic about link information of users establish friendships and ownership of objects and labels,this paper presents a social networks recommender method based on non-negative multiple matrix factorization,considering the multi-source information of the friendship between users,characteristic information of users and resources,as well as the ownership of resources and tags.On the basis of the non-negative multiple matrix factorization recommender algorithm,this paper discusses and analyzes the matrix of user friendship,and proposes a non-negative multiple matrix factorization algorithm to improve the user friendship matrix.Finally,based on the Sina micro-blog data set,the resource content recommendation experiments are carried out.The results show that the proposed method has better performance in improving the recommendation effect;based on the Last.fm and Delicious data sets,resource recommender contrast experiments of improved friendship matrix are carried out,discussed the different link prediction method for the effect of different social network recommendation,recommending friends to users according to the results of the resource recommendation.Based on the analysis of the experimental data,the proposed recommender method based on the non-negative multiple matrix factorization can effectively utilize the multi-source information in the social network and significantly improve the accuracy of recommendation.This method can not only dig out the tags of resources,but also mine the relationships among users,and can also extend to the recommendation of friends and labels to the user.