A Study On Evaluating User Influence Based On The Relationship And Behavior In Microblog

After the advent of Web 2.0 era,people had be accustomed to consuming most of their fragmentary time on Microblog.Nowadays,Microblog has been the main platforms for online marketing.Meanwhile,it has received much attention in the academic world.In the case of the data mining,there is a focus on measuring User' influence in Microblog.There are three main research directions for analyzing user influence: research on propagation models,optimization of propagation paths,and learning of impact metrics.This paper aims to analyze the influence factors of Microblog users based on user relationship and behavior,and construct an influence evaluation model.This paper analyzes the influence factors of Microblog data and summarizes it into three parts: trust,activity and interaction.The degree of trust includes user static attributes such as the user's fan volume and class;the activity is mainly the number of microblogs published,which is the dynamic behavior characteristics of users;the interaction force includes the number of mentions,comments,forwards,and likes,and is interaction behavioral characteristics.This paper uses the Delphi method to collect expert opinions,analyzes the opinions of experts through the analytic hierarchy process,and finally determines the weights of each indicator in the impact assessment system,and compares the actual data with the validated data to verify whether the weights are reasonable and effective.The simple quantitative analysis of user influence is not applicable to microblogs with complex relationships,so this paper uses PageRank to analyze the user influence in the microblog network structure.This paper analyzes Microblog 's concern network,and forwards and mentions two behavioral networks.It improves the definition of transfer matrix of PageRank algorithm to make PageRank more suitable for Weibo user impact analysis,and integrates microblogging macro into three networks.Indicators(such as fan volume,forwarding volume)calculate the user's initial impact to reduce the interference of the data as a network “snapshot” to the impact calculation,so that the measurement results are no longer one-sided and outdated.Based on the behavior analysis,the influence will be strongly interfered by the water army.By adding the Weibo level,the ratio of forwarding and forwarding,etc.to the algorithm,the “water army” indicator will reduce the initial influence of the “water army”.Factors such as activity,comment,and praise are occasionally strong.Therefore,these single factor influence calculation methods use the averaging method and the normalization method.The MBRank algorithm matches the influence of each factor to the weight,and the comprehensive measure calculates the user influence.Considering the huge user volume and complicated interpersonal relationship of Weibo,the three network single factor influence calculation iteration matrix multiplication times,but it does not affect each other,using the pseudo distributed model of Hadoop platform,supplemented by HDFS and MapReduce implements the MBRank algorithm efficiently based on the results of Weibo data calculation.It has been proved by experiments that the MBRank algorithm is more comprehensive,more accurate and more reasonable than the quantitative analysis method and the PageRank algorithm.