Research Of Personalized Tag Recommendation Algorithm Based On Improved Tensor Decomposition

With the rapid development of social networking sites,social tagging has gradually become a research direction that people are very interested in Web2.0.Social networking sites are becoming more and more popular on the Internet because of the simplicity of using open tags to sort and retrieve content.Social annotation helps users to annotate and categorize items.More and more users provide information about users themselves through social tags.Social tags describe both the semantics of the items and the preferences of users.However,the current problem with the tag recommendation system is that the user manually inputs fewer tags,resulting in a large number of data sparseness problems in the social tagging system and a large number of ambiguity problems and semantically ambiguous tags.Traditional tag recommendation algorithms usually decompose three-dimensional data {users,items,tags} in social networking sites into three groups of two-dimensional relationships {users,items},{users,tags} and {items,tags} for analysis.Therefore,the traditional tag recommendation algorithm will lose some of the information hidden in the three-dimensional data,resulting in the structure of the high-dimensional model data is destroyed and the accuracy of the tag recommendation is reduced.The tensor model can completely represent all the information contained in the high-dimensional data.In this paper introduces a three-dimensional tensor model,studies the tensor decomposition and its application in the tag recommendation system,and uses the three dimensions of the three-dimensional tensor to describe the three types of entity users,items and tags in the social tagging system.The method of singular value decomposition of order reduces the tensor dimension and further improves the accuracy of the recommendation system.In order to improve the recommendation accuracy,efficiency and quality of the tag recommendation system,this paper proposes an improved tensor decomposition recommendation algorithm.The recommendation algorithm first performs similarity calculations for users,classifies users with similar characteristics to ensure that the data has initial aggregation,optimizes the weights,and then applies high-order singular value decomposition to establish a three-dimensional tensor model for the grouped data.Experiments show that this algorithm has an improved effect on solving sparsity problems in recommendation systems and improves the accuracy of the recommendation results.