Research On Tag Tensor Based Personalized Recommendation Method

Personalized recommendation system is an information filtering applications. Itmainly refers to analysing the interest based on behavioral characteristics of thesystem user, and making recommendations to the user with more interested in contentrelevant. Existing recommendation system is generally used for two-dimensionalrelational data between users and items to make recommendations. With thedevelopment of Web2.0, user autonomy on the Internet is growing. Social taggingsystem for its freedom, flexible and open organization are recognized by the majorityof Internet users, and has been widely used, its rich User-Item-Tag ternary contains ahuge commercial value. To take full advantage of the value of social tagging systemsto improve service levels recommended, we proposed Tag Tensor Block ComponentDecompositions(TTBCD) personalized recommendation method.To deal with sparse data for the current tagging system, the inherentcharacteristics associated with complex, we use tensor data on social tagging systemsmodeling, and get the tag tensor models. Through core item-based clustering, the tagtensor is clustered into sub-tensor with different item theme features, and the resultingsub-tensor is decomposed by using Tucker decomposition method, and getapproximate tensor. On this basis, we propose a personalized recommendationmethod based on tag tensor block decomposition. For a given user, first obtain theuser associated tensor decomposition model with the different sub-theme of item, anditem extraction associated with the user. Then through computing item and userconcerned by the theme of matching method, we get Top_N personalized itemrecommendation results. And experimental results show that the Tag Tensor BlockComponent Decompositions personalized recommendation method can handleefficiently with the tag data without losing the intrinsic relationship, and producemore accurate and personalized recommendations results.Firstly, we studied the current popular social tagging system carefully, andanalyzed the characteristics of tag data, as well as the effect of the recommendation system based on tag data. Then we presented the tag data tensor modeling, and themodel of tensor block decomposition. On the basis of tensor block decomposition, weproposed Tag Tensor Block Component Decompositions personalizedrecommendation method, and given a recommendation system framework. Thencomparison test on MovieLens, LibraryThing, and several other commonly usedpublic experimental data sets. Finally, we summarize the work of this paper anddiscuss future directions of research.