Research On Collaborative Filtering Recommender Algorithm Based On Social Interest Clustering

With the development and upgrading of internet software and hardware, itprovides an excellent platform for the boom of e-commerce. Meanwhile, it lead toe-commerce users facing more and more vast amounts of information, includingproduct information, and custom’s review and rating, which makes users obtaincommodity information which meet their needs even more difficult, so the e-commerce user face serious "information overload" and "select barriers" problem.Search technology arises, it can solve those problems in certain extent, but theinformation is obtained through search engine is no-personalized, and search enginecannot meet the user’s personalized needs in the e-commerce website, so the userneeds e-commerce website recommend some commodity help them make a decisionaccording to their preference in the website.As the emerging of recommender technologies, especially CollaborativeFiltering which is one of the most widely used recommender techniques.Collaborative Filtering recommendation techniques make recommendation to userbased on their ratings information of the commodities, so it ignores some importantof the UGC (user generated content), such as: tags, reviews, and other vitalinformation, however, current research already exists use the tag semantics analysisand community mining techniques to analyze problems at the same time. This studyaims to make the comprehensive utilization of rating and tag information to makepersonalized recommendations to the user, and takes into account the usercommunity information in real life. So the research calculates user similarity usingrating from two different angles and use tag information to mining e-commerce usersocial interests, then cluster user through user social interests, at last, usecollaborative filtering techniques based on the cluster result to makerecommendation. In theory, the model can improve the recommendation accuracy toa certain extent.This research uses the empirical research method, verifies and compares themodel with other traditional recommender techniques on the Movielens data setusing MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) toevaluate recommendation quality. At last, this paper analyses experiments resultscomprehensively, make conclusion of this research and give the future work.