Hybrid Personalized Recommendation Method Based On Social User Tag

With the development of the Internet and information technology, numerous socialnetworks have emerged, such as the common Renren and Facebook, the latest newssocial, workplace social network, and so on, the network life gradually becomes social.Besides, network information increases sharply, how to find the interested resources invast amounts of information has become increasingly important. However, thepersonalized recommendation system, which not only can improve the customersatisfaction and loyalty, but also can enhance income of the enterprise, could solve thisproblem. Therefore, the personalized recommendation system has become a hot topic inthe field of intelligent retrieval recently.Collaborative filtering is the most widely used and successful recommendationtechnology. It only dependents on the user rating matrix of the project, and has goodapplicability for various specific applications. However, it also has some problems, suchas the recommendation accuracy, cold start and coverage problem. In order to solvethese problems, inspired by the previous work and by considering users’ features ofsocial network, a method of hybrid personalized recommendation based on social usertag and collaborative filtering is proposed.In this paper, the main research work and achievements are as follows:①In view of the collaborative filtering accuracy, cold start and coverage problem,a hybrid method based on social user tag and collaborative filtering recommendation isput forward. First of all, through traditional user-based collaborative filteringrecommendation algorithm can generate recommendations, then use users personalizedtag and user relationship in the social network to recommend, finally combine these tworesults for hybrid recommendation;②In the hybrid method by using social relations and the user personalized tags,and structural similarity measure method is adopted to implement a recommendationmethod based on social user tag;③By using weighted combination, social network and collaborative filteringrecommendation are mixed to generate the final result.Finally, the proposed approach of this paper is evaluated by general data sets andcold start data sets, respectively. Through the experiments on the accuracy, recall andthe coverage, it shows that this method could improve the accuracy and coverage, and also has a better effect on the problem of cold start.