| As the explosive growth of the Internet and the coming age of web2.0, the applicationsbased on Internet grow quickly, the representatives are the e-commerce system and manykinds of virtual social platforms. Amazon, Douban, Facebook, Flickr have attracted a lot ofusers on the Internet. A large number of products, services and kinds of large-scaleinformation appear on the Internet. In order to help user to find their interested products,recommendation system has been largely used on e-commerce site and social network,recently the study of recommendation system has become a hot field.Recommendation system is a personal information filtering skill and responsible forrecommending interested products or services to users. Traditional recommendation systemsmake use of the attributes of users or products to carry out content-based recommendation,ormake use of similarities among users or products to carry out collaborative filteringrecommendation. Since many public taxonomy(folksonomy) applications allows users to usetags to classify and label products, This has provide more information for recommendationalgorithms. So, how to use the tag information, how to build a model about users,products,tags and attributes are becoming a hot spot in recommendation.This paper purpose a recommendation algorithm named "based on weight withintripartite graph diffusion of personalized".(short for “TGD recommendation algorithm”).TGDbuilds the model with triples("user-item-tag") using the tripartite graph and introducing theweight of user’s importance to initialize the tripartite graph.It presents the personalizedrecommendation. Splitting the tripartite graph model to binary graphs, adding the weight onthe edges and making diffusion. Finally integrating the final weight of diffusion to get thetop-n items.Our experimental evaluation on three different datasets which are MovieLens,Flickr,Delicious. The experiments shows that TGD play better than the based tripartite graphalgorithm on accuracy. |
PDF Full Text Request