Tag Recommendation Algorithm Based On The Tensor And The Tripartite Graph Model

With the rapid development of information technology and the Internet,Information overload problem has seriously affected the user experience degrees.Information producers and consumers as the two camps of Internet will face great challenge.As the information consumers,how to quickly find information they are interested in,As information producers,how to make information got the attention of the masses of users,all of this will be a very difficult problem to them.At this point,the recommendation system arises at the historic moment,solve the problem very nice.Recommendation system can contact users and information,On the one hand,can help users find valuable information on their own,on the other hand can let information show in front of the users who interested in it,So as to realize the win-win situation of information producers and consumers.But with the expansion of the Internet category,recommendation system also gradually detailing,According to the emphasis of generate data from different Internet application,therewith generate different recommendation algorithms,for example,According to user's behavior data,the collaborative filtering algorithm based on user or item was put forward.According to user's tag data,the tag recommendation algorithm based on the model or drawing was put forward,According to the context information about time or place,the recommendation algorithm based on time context or location was put forward.According to the social network data,the recommendation algorithm based on social network was put forward.This paper focuses on study tag recommendation algorithm,in social tag recommendation system,the user tagging information for the item which they have browsed in the form of a "tag",to obtain the accurate keywords description of the item,so as to improve the performance of the recommendation system effectively,In this paper,we put forward two kinds of tag recommendation algorithm.the tag recommendation algorithm based on tensor decomposition and the tag recommendation algorithm based on the time-weighted tripartite graph.The emphasis of this two algorithms are different,the tensor decomposition algorithm using tensor model to get the potential semantic association among users,items and tags,under the condition of offline calculates the connection weights,and under the condition of online to recommend tags to users.And the time-weighted tripartite grap algorithm is to use the tripartite grap model and add the time attenuation factor to realize a efficient on-line real-time performence of tag recommendation.The tag recommendation algorithm based on the tensor decomposition introduce a tensor model,using the third-order tensor to describe the three types of entities of social tag recommendation system:users,items and tags.based on the tagging metadata to construct initial tensor and using the tensor reduction and the Higher Order Singular Value Decomposition(HOSVD)method to reduce the dimension of tensor,at the same time to realize the analysis of potential semantic association between three types of entities.to further improve the accuracy of tag recommendation system.The tag recommendation algorithm based on the time-weighted tripartite graph introduce a weighted,undirected tripartite graph model,Firstly combine the BM25 weighing scheme and tag time attenuation factor to calculate the link weight of tripartite graph,and generate the adjacency matrix for the folksonomy graph.finally applying the Katz measure,a path-ensemble based proximity measure,to predict both the path weight between user and tag,and the path weight between item and tag.Thereafter,We calculates a ranking score of tag based on such predicted path weights and thus generates a list of ranked tags suited to user for item.Theoretical analysis and experimental results show that the tag recommendation algorithm based on tensor decomposition and tag recommendation algorithm based on the time-weighted tripartite graph on the accuracy and recall evaluation index has obvious improvement in performance than typical tag recommendation algorithm.