Microblog User Preference Analysis And Modeling

As a rapidly developing social network application in recent years,microblog has become one of the important tools for people to communicate and gain information.Microblog content contains the user preference.The user preference information implied in Weibo plays an important role in modeling users,recommending content,forecasting hot events and providing users with personalized services.However,the user preference hidden in the microblog text is hard to be directly obtained and described,therefore how to accurately mine and describe the information becomes an urgent problem.This dissertation uses Sina Weibo data as the object to analyze Weibo user preferences description and modeling methods,and verify the methods through experiments.The main work of the dissertation is as follows:(1)This dissertation considers the behavioral characteristics of Sina Weibo users and the fact that user preference will be shifted over time to improve the method of keyword weight calculation.Then it introduce the keyword weight into graph model as the node’s weight,and use the improved graph model to extract words that contain user preference.Experimental results show that,compared with the WTM algorithm,the proposed algorithm of interest keyword extraction improves the accuracy by 2.97%and improves the recall rate by 2.67%,and it also performs better than other classic methods.(2)Based on the assumption that users are likely to show more intense emotions for more interesting things,this dissertation proposes a sentiment analysis method based on dependency grammar and semantic dependency,and introduces the result of sentiment analysis as a preference factor into user preference analysis of the process.Experiment results of sentiment analysis show the proposed method performs better than the classic methods,it improves the precision,recall,F-measure by 2.03%,2.98%,2.53%.The results of user preference modeling experiments based on micro-blog data also show that the user preference modeling method combined with emotional analysis is more applicable.