Research On Personalized Microblogs Corpus Recommendation Based On Users Interests

With the rapid development of mobile intelligence and internet, people marched from thetimes lacking of information over the years to the era of information overload. As a newsocial self-media platform, micro-blogging experience a exponentially grow of its users inrecent years, generating a large number of UGC (User Generating Content) information everyday, how to tap the information of users‘personal interest and establish user interests model,find out users‘most interested information from the mass of information and recommended tothe users appeared very important.This paper focuses on the modeling of micro-blogging users‘interests and therecommendation of personalized information, the main contents of this paper is:(1) The traditional Vector Space Model and the TF-IDF Method does not consider thepresence of semantic information and has problems such as sparse user characteristics of high–dimensional. And the Latent Dirichlet Allocation Model (LDA) based on document-levelco-occurrence of words is not applicable for topic mining and user interest modeling in shorttext such as micro-blogging. In view of this, this paper introduces the theme model BTM(Biterm Topic Model) for short text to excavate user’s personal interests, while taking intoaccount the variability and different levels of activity of user interest, proposes a timewindow based dynamic user interests model.(2) On the basis of the model of the user’s personal interests, against the micro-bloggingusers‘user-listen-to-list information overload problem, this paper uses the rich featureinformation of micro-blogging, presented a recommendation model which integrated threemain characteristics, the quality of micro-blogging itself, user’s personal interests and socialinterests, and introduces the idea of collaborative filtering in the model. As for additionalinformation which micro-blogging users actively acquired (non-user-listen-to-listinformation), this paper designs a theme based information recommendation application.(3) This paper obtains experimental data through micro-blogging API and crawlingmachines, proves the user’s interest constructed by BTM model is superior to LDA model andTF-IDF model in the performance of information recommendation and consider thevariability of user interest to further improve the recommend effect. Among the three main influencing factors, performance of the combination of collaborative filtering idea anduser’s personal interest characteristics recommendation is optimal, followed by users’ socialinterest characteristics, the worst is microblogging itself’s quality characteristics;The recommendation model is proposed from the modeling of user interests, combinesdifferent characteristics to build the recommendation model under different scenes, anyinformation recommendation problem on UGC platforms can extend and use the study andachievement of this paper.