Research Of The Recommendation Algorithms Based On The Weight And Time Factor Of Tags

With the development of web2.0, douban, Youtube, Facebook and many other socialnetworking sites rise gradually, users on Internet are no longer just the information readers,but become the information publisher. Users can upload resources and are allowed to useany texts called tagging label resources. Tagging system have become the important roleof Internet recommendation system. However, because tagging have UGC (User GenerateContent) and Free Tagging features, making the tagging system face problems ofinformation overload and existing garbage labels, and those problems have an impact onthe precision and the degree of recommendation system. How to use tagging system inrecommendation system more reasonably is the key to solve the above problems. Themain research in this paper are as follows.Firstly, an in-depth analysis of the characteristics of tagging system is carried out, aswell as the role that tagging system plays in the recommendation system. Then the usermodel based on tags is introduced, and several common recommendation techniques andapplication methods based on tags are outlined.Secondly, according to the recommendation algorithm based on graph, the personalrecommendation is improved by using two weights of tag to improve the diffusionalgorithm based on tripartite graphs. Another new weighted tripartite graph model basedon tags is proposed. The application of diffusion algorithm in this model, gives fullconsideration to the bridge role of tags. And comparison and analysis with the diffusionalgorithm based on original tripartite graphs is carried out.Then, taking into account the time utility of the recommendation system, weproposed a method that combined time weight of tags and frequency of use tags, toimprove user-item rating matrix. This matrix looks the good impression of the user asratings of resources, to better reflect users’ interests with time and help to obtain the recentinterests of users. The new rating matrix is used in collaborative filtering algorithm tomake recommendations for the user, and make a comparison with the original method thatonly use tags’ frequency for the rating matrix. Finally, the experimental evaluations of two kinds of diffusion algorithm on tripartitegraphs and a collaborative filtering algorithm based on the time weight of tags is carriedout, then analysis of accuracy and diversity is made.