Research On Prediction Of Microblog Propagation Effect Based On Analysis Of Users’ Roles

Microblog is a platform of information accessing, sharing and diffusion based on user relationship. As a new kind of media, it changes the message traditional spreading way of media and arouses a revolution in the field of information propagation. The research of microblog propagation effect based on analysis of user ro les is of great significance to improve marketing efficiency and supervise public opinions. However, the exiting research lacks consideration of the users’ difference, and users, provider and communication subject of information in the microblog network, have an enormous impact on the effect of information propagation.This paper launches the research of user roles from user influence and information propagation effect with the analysis the users who affect the information propagation. Then, we take the retweet scale and diffusion depth of message as the eva luation criterions of propagation effect. The following contents are our main work:1. We propose a method of evaluating microblog users’ influence based on comprehensive analysis of user behavior. To overcome the disadvantage of current evaluation method of users’ influence without considering user link relations and behaviors comprehensively, we make use of user activity and mutual actions to filter the user relation. Then t hrough comprehensive analysis of retweet, comment, mention and statistical analysis of different contribution of different behaviors on influence, a Page Rank algorithm based on multi-behavior weight amalgamation distribution is proposed to evaluate spreading influence quantitatively. Experimental results based on Sina microblog data sho w that this algorithm is more accurate and effective than other traditional methods in finding influencing users.2. We propose a role analytical model of users based on network structure features. To overcome the disadvantage of current research without characterizing the function of users in the information propagation process adequately, we introduce the features of network structure and neighbor nodes based on extraction of characters of users’ own attributes and improve k-means algorithm combining with the feature of information retweeting diffusion. Eventually, users are divided into three sorts: commom users, bridge users and core users. Experimental results on Sina micro-blog dataset show that the proposed model can better differentiate users according to information propagation3. We propose a predictive model of propagation effect based on users’ re tweeting behaviors. To overcome the disadvantage of current prediction research of propagation effect without considering user difference comprehensively, we propose a predictive probabilistic model of retweet behavior based on logistic regression algorithm and relative features extracted from users, relationship and contents which include role analysis relative features. Based on this model we propose a method of predicting scale and depth of retweeting considering the character of information disseminating along users. Experimental results on Sina microblog dataset indicate that these models all have good predictive results. Finally we conduct information dissemination experiments on three types of users with the application of SIR model and r etweet probability. The simulation results illustrate that the propagation effect of core users is best, followed by bridge users and common users.