Social Tagging Semantic Normalization And Its Application On The Recommendation System

With the rapid development of Internet and Web2.0 technology,the application of social label system has become more and more widely,and the traditional user access to information through the browser is different,the user can freely marked according to their own interests interested Of the resources,the user is not only the user information is the information provider.Because of the rich personalized information and semantic information,the label is more and more applied to the recommendation system.However,the user rating data in the recommendation algorithm is sparse and the recommendation quality is not high,and the user's label information has a certain semantic ambiguity and other problems can not be effectively applied in the proposed algorithm.To this end,this thesis proposes a social label semantic normalization method and applies it to the recommendation algorithm.Specific work is as follows:First of all,through the study of the social label system,the short and fragmentary nature of the label itself is analyzed,and the semantic association between the external lexicon based on English Wikipedia and Word2 vec semantic model is proposed,which avoids complex semantics Analysis process.And proposes a label normalization method based on network segmentation clustering.It is considered that user-defined random labels can be replaced by normal tags whose resources core is similar to the random labels,which is convenient for the construction of user label interest model.Secondly,the existing interest weighting method for label interest model is improved,and the information entropy calculation method based on measuring the uncertainty degree of user labeling score is proposed,and the user's normalized label interest model is applied to collaborative filtering algorithm.The construction of user interest model can make full use of the rich personalized information,and effectively reduce the vector dimension of user interest model.The feasibility and effectiveness of the collaborative filtering algorithm are tested by using MovieLens and instance data from Delicious.Compared with the previous tag-based recommendation algorithms,this method can solve the problem of label semantics better,and can effectively alleviate data sparsity and recommend cold start and improve the quality of recommendation.