Extracting And Dynamic Modeling Of User Interest On Microblog

With the development of social media,the Internet has changed people’s life and interpersonal communication.Microblog shows Strong momentum of growth by its performance of original,timeliness and grass-roots and it has enriched the content of the Internet.People on Micoblog can follow the personal or public account according to their own interests,and they can receive lots of information on microblog.Micoblog provides lots of topics and content for users,people can get information based on their own interests.Therefore,microblog is content platform besides socialization platform which is led by interest.In this condition,in order to improve user’s experience,analyse user’s behavior trace and personalized recommendation,it is important to extract user’s interests and dynamic modeling user’s behavior.Our work mainly focuses on two aspects,the first part is mining user’s interests by analyses their text content of micoblog,making preparation for dynamic modeling.The second part is achieving dynamic behavior modeling on time base by researching people’s microblog and predicting user’s action and making personalized recommendation based on the model.In the first part,we adopt unsupervised learning to extract user’s interests,because user’s interests are very personalized and uncertain.First,we consider microblog content as short text and cut text into Chinese terms.Then,we reference similarity to represent distance between different texts.Next,we use LDA(Latent Dirichlet Allocation)topic model and k-means clustering method to analyses user’s interests.Finally,we find the user’s interests tag based on above-mentioned.When we model user’s behavior based on their interests,we target interest-related microblog as the object of the research.First we sort the microblog based on time,then we build a dynamic model on user’s interests transform and simulate the user’s attention’s transform among interests.Then the model can help to predict and recommend the content to users and we use recommendation precise to estimate the dynamic model,and the precise of recommendation reaches 78%.