User Influence Model Based On Comment Authentication And Interaction Behavior Analysis

Social media likes a revolution because it has brought new trends in communication,shopping,work and other areas of life,thus changing many aspects of people's lifestyles.On microblog websites,more and more users interact with each other every day,and various ideas emerge,spread and collide rapidly.In this context,social media can be seen as one of the most important sources of information influencing public opinion.As the representative of such a new social media platform in China,Sina Weibo attracts many users to share experience and express opinions on the platform with its extremely high openness,immediacy and sharing.It now has 465 million monthly active users.Weibo has also become an ideal platform for conducting electronic public opinion surveys on products,services and public figures,and an important front for advertising marketing,public opinion monitoring and information dissemination.In order to promote sales,weibo big VS will buy online water armies and forge fake comments in order to get higher advertising prices.Nowadays,social networking platforms are full of false information,and the online environment is rapidly deteriorating.On the other hand,the research of fake comment recognition technology can improve users' online social experience;on the other hand,it can improve the accuracy of users' influence and the mining efficiency of high influencers.In hot events,public opinion should be guided to suppress the spread of rumors and prevent the spread of untrue information from causing adverse effects on society.Therefore,this dissertation takes Weibo,a social network platform,as the research object to discuss how to effectively remove the interference of fake comments and build a user influence model.To sum up,the research content of this dissertation mainly includes two parts.In the first part,considering that fake comments will reduce the accuracy of user influence calculation,a Mixed Feature Approach(MFA)based on content and behavior is proposed from the status and behavior of comment users,and 84 content-based and behavior-based features are extracted.Three machine learning classifiers were used to detect fake reviews.After experiments on two data sets,the multi-layer perceptron with the best classification effect is finally selected as the classifier of fake weibo comments,which is used to filter fake weibo comments.The features were divided into 6 groups,and different combinations of features were used for experiments to test the effect of extracted features on classification.The experiment proved that when all features were used,the detection accuracy was the highest,and all extracted features had positive effects on classification.The second part explores the model of user influence.This dissertation firstly reviews and analyzes the research status of user influence at home and abroad,summarizes the development status of user influence model,and specifically describes the improved algorithm based on Page Rank by various researchers.This dissertation proposes a quantification algorithm of microblog user influence based on user interaction behavior.Social link relation,interaction data set and user comment data set are extracted by using microblog data.Then,aiming at the problem of incomplete social link network caused by anti-crawl setting of micro-blog,this dissertation proposes to adjust network structure by link prediction.LDA algorithm is adopted to extract the comment subject words of each user according to the content published by microblog users,and calculate the topic similarity of users.Finally,a modified Page Rank algorithm based on user interaction behavior is proposed to calculate user influence by combining the normalized topic similarity as link weight and the Page Rank algorithm which considers user direct influence and potential influence.Finally,the accuracy rate and recall rate of high influencers mining are taken as evaluation indexes,and the effectiveness of the model is verified by comparing with the user influence model,Leader Rank model and user popularity without filtering fake comments.Through experimental analysis,the user influence model proposed in this dissertation has a good performance,enriches the research content of user influence,and can provide reference for the research of user influence in social networks.