Micro-blog attracts a number of users immediately because it’s short, instant, andconvenient characteristics and produces a large number of information flow by rapidtransmission and spread of fissile transmits and comments. An accurate and efficientmicro-blog user’s interest recommendation method needs to be constructed. Thereforethe micro-blog topics that users are interested in can be discovered. Micro-blog userinterest model is one of the most important elements for user’s interestrecommendation. So the study of micro-blog user interest model construction iscrucial for the development of micro-blog website.Aiming at micro-blog user interest information acquisition, micro-blog userinterest model building and establishing of recommended method, this thesis regardsthe construction of Micro-blog user interest model and the recommendation methodas the research background. A new micro-blog user interest model constructionmethod is presented and the recommended method is proposed in order to realizeaccurate and efficient microblog topics recommendation for micro-blog users.Thisthesis takes sina micro-blog as data source, mainly concerns the following issues:(1) All kinds of information in micro-blog, micro-blog users’behavior as well asthe relationship between micro-blog users’ interest are analyzed, and the properinformation as the source of user’s interest is selected to describe accurate informationof users’interest.(2) A non-negative matrix factorization based on the term correlation andnormalized cut weighting for micro-blog user interest model is proposed. First, a termcorrelation matrix using term distribution context is constructed to better explainsimilarities of terms, and then Ncut-weighted Non-negative Matrix Factorizationmethod is presented to obtain the matrix of user-topic matrix, which shows theclustering results of user interest. Experiments show that, this method can effectivelymicro-blog topic clustering to support micro-blog user interest model.(3) A microblog recommendation method based on user interest model andConversation extraction is proposed. First, a Ncut-weighted Non-negative MatrixFactorization is applied to obtain user-interest matrix. And then Single-Pass clustering based on conversation extraction using frequency and correlation is used to betterexplain similarities of terms, Experiments show that, this method can effectivelymicro-blog topic clustering to support micro-blog user interest model.Experiment results show that the models presented in this thesis can effectivelyrepresent user interest and improve accuracy of the recommendation result. |