Weibo User Interest Modeling Based On Multi Dimensional Feature

Internet has experienced rapid growth in recent years,and with a variety of intelligent mobile devices gaining popularity,social networks such as Twitter and Weibo are still on the rise.As the carrier of network information,social network itself is fast,convenient,short and flexible.It connects a large user base through“following" and "followers",allowing more people to participate in the dissemination of information through"forward" and "comments".Although social network has brought a huge impact on the way people communicate and access information,it also has its own limitation which is "information overload"."With too complicated information,users sometimes can not effectively get the useful ones,which is not beneficial for the spread of information.Solving this problem requires a social networking platform to understand more about the user,through modeling the user's interest and preferences in an accurate and comprehensive manner,so as to lay a solid foundation for a variety of personalized services.Based on the context above,this paper focuses on Weibo users to study the multi-dimensional hierarchical user modeling method,where multi-dimensional refers to cover as much features as possible to describe the characteristics of users,hierarchical refers to these features carding relations,forming a hierarchical structure,reducing coupling,so that the model is scalable.This paper mainly completed the works as follows:1.Design of Weibo catcher system.Achieve effective catching,processing and storing procedure.2.Distinguish user nodes.Including finding important user nodes using Page-rank method and deciding the active user nodes calculating the active degree.3.Model short articles.To overcome the problem of short articles,irregular words and noise,this paper introduces subject model to train paragraph vectors containing subject information and using continuous vectors to represent weibos.4.Construct multi-dimentional hierarchical user model.Construct each part of the model separately,summate the weighted similarities of each part during calculating finally experiment the model in real life scenario of user friend recommendation.