The Tag-based Personalized Recommender System Reaserch

With the new generation of Web2.0, Tag, a new and particular technology whichcan implement personalized classification of the Web items, points a new way to thepersonality recommender system. Web users tagged Web items by tags, which notonly can help users organize and search the Web items, but also can find interestingWeb items and similar neighbors by analysing users’ behavior and preferences. Therating of item by user can reflect users’ preference directly, and the average rating ofitem can reflect the popularity of the item. So, we combine the rating with tag-basedrecommender system, construct a tag-rating-based recommender system (TRRS) toanalysis the effect, apply and influence of the rating.In this paper, we use the users’ rating, information retrieval and set theorymethod, propose an adaptive method for the TRRS and a similarity method for theTRRS. The main work is as follows:(1) An adaptive method for the TRRS:The qualitative measure of users’ rating makes use of the rating preferencerelation on (user, item, tag). We reflect the user-item-tag (UIT)3-dimentional space tothe user-item (UI)2-dimentional space to propose users’ rating preference and items’rating preference. For these two rating preferences, we construct the TRRS networkconfiguration to generate UI recommendation and IU recommendation respectively.By the use of an adaptive method, we construct the user interest model and implementtwo-way recommendation with recommend item category to users and recommendtags to items.(2) A similarity method for the TRRS:By the set theory method, the qualitative measure of similarity makes use of therating preference relation on (user, item, tag), and the quantitative measure makes useof relation matrix reflection on (user, item, tag). We extend the I-basedrecommendation algorithm only based on the item influence set to the UIT-basedrecommendation algorithm based on the user, item and tag influence set. The k nearestneighbors (kNN) and revers k’ nearest neighbors (RkNN) combines with thesimilarity methods are used to generate recommendations. We proposed4formulas topredict rating for users.The experiment results show that the combination of users’ rating with Tag-based recommendation system makes the positive function for improving theaccuracy. UIT-based recommendation algorithm has better recommendation efficiencyto the I-based recommendation algorithm. The combination of user, item and tag canimprove the accuracy of recommender system.