Research On Modeling Of Micro-blogger’s Interest And Its Changing Pattern

The development of social media has made Internet users can receive andpublish new information anytime, anywhere. As a new form of broadcast, microblogextends traditional social media to a platform to receive information and operatemarketing at the same time. On the one hand, this kind of platform enablesmicroblogging users see information from various sources, on the other hand,businesses can easily conduct product promotion and publicity through this platform.While this propagation mechanism can bring many benefits, but for the users,"information overload" makes them have to read large number of micro-blog contentto get the information they really care; for businesses, pushing commodity-relatedcontent equally to all users not only highly costly, but also easily produce negativeeffects. This paper targets microblogging users, proposes a interest model todetermine the expression of intereste, calculation methods and forms of storage,based on the micro-blog users interest model, Markov model is made to picture theinterest’s changing pattern.Firstly, by reviewing and summarizing the domestic and foreign scholars’researches on intereste modeling and interests drift, we give definition of weibousers’ interest and its changing pattern. By analyzing different information used andapplication range of different expressions during interest modeling, we determinethe information and forms of expression of interest used in this research, andcombining some of the classic algorithm in text processing, we calculate the degreeof interest. Secondly, we determined different period of interest and vector of users’interest in each period, and variation of the vector of interest are used to formcorresponding statistical Markov models. Finally, we predict user interest vectorsand evaluate them to determine the best length of the period, using the predicteduser interest vector to do content-based recommendation and collaborative filteringrecommendation.In this study, we use real content from Sina Weibo microblog as the experimentdata, and propose the Pearson correlation coefficient and rank accuracy to measurethe quality of the dynamic model to determine the appropriate length of period usedwhile modeling the users. This research has theoretical and practical significance onpersonalized service and recommendation systems.