Research And Implementation Of Recommendation Method In Social Tagging System

In the rapid development of the Internet era,the social tagging systems like DouBan,CiteULike have become an integral part of people's life.These systems bring people convenient knowledge sharing,meanwhile cause the information overload.The recommendation method can intelligently offer user interested information and potential friends.Hence,the recommendation method in social tagging system is getting more and more attention.However,compared to traditional recommendation system,social tagging system has not only users and resource,but also a new evaluation criterion,tag,which leads to the different challenges for the recommendation in social tagging system.They can be categorized into three aspects.At first,users in the social tagging system can use arbitrary words as tag to describe resource when they post or comment on resource.The greatly freedom and openness results to the low quality of tag.Secondly,the cold start problem of posting new resource and the data sparsity of user's tag also need to be tackled.Thirdly,the traditional recommendation method uses unified resource model which neglects the fact that resource like document or picture has potential category.Therefore,to solve the problems of low tag quality,data sparsity,cold start and unified model applicability for traditional recommendation method,this paper proposed TECA(Tag Expansion and Content Analysis),a method based on tag expansion and content analysis for tag and user recommendation of new resource.TECA will do the resource classification training and build model for each class of resources which will avoid the problem of unified model applicability.When building user model,TECA will use neighbor's tags to expand the tags of target user for alleviating data sparsity.And to ensure the quality of tags,it use the historical tags and topic word of the resource to recommend tags.What's more,TECA analyses the content of resource and use the semantic topics to compute similarity which can overcome the cold start.This paper conducted recommendation experiments on the dataset of CiteULike and made comparisons with traditional collaborative filtering recommendation method.The experiments show TECA achieves better tag and user recommendation performance than traditional collaborative filtering recommendation method and prove the improvement of model applicability and data sparsity for TECA.In addition,we took experiments to find the proper parameters for the TECA.