Research On Opinion Leader Mining And Its Information Dissemination Model Based On Two Stages In Social Networks

At present,the social network becomes an important platform for people to exchange information.Among the users involved in social network activities,we call a small portion of them opinion leaders as they are sought after and accepted by more others.During the process of information interaction,opinion leaders have significant effects on people in terms of thoughts,feelings and actions.And they play important roles in researches concerning the information dissemination,in public opinion guiding and supervising.Thus,mining opinion leaders in social network has become a key part of social network analysis.Though opinion leaders account for a lower proportion in social network,the majority of mining methods evaluate the probability of being an opinion leader for each user,which leads to a high calculation complexity.In addition,opinion leaders are easier to spread information online than that of ordinary people in the real world.But many researchers often neglect the effects of user's role for information spreading during the research of information propagation model,which causes deviation of the information diffusion prediction from the practice.Based on above two problems,this paper introduced a two-stage opinion leader mining algorithm,as well as an linear threshold model based on user's role.The main work done includes:(1)We introduced a two-stage opinion leader mining method.The proposed algorithm divided opinion leader identification into two stages: Clustering and Ranking.In which,Clustering stage clustered users in social networks using K-means algorithm,according to the topological information that can fully exhibit opinion leaders.Then it chose clusters that meet the conditions of opinion leaders into opinion leader candidates,which aims to reduce the scale of data calculation.In Ranking Stage,after analyzing opinion leader candidates' historical behavior data,user leadership was calculated from the perspective user activeness,user influence and spread centrality to identify opinion leaders.(2)We introduced an linear threshold model based on user's role.Firstly,the proposed model determined the influence weight between each pair of active user and inactive user,according to their different user roles.Simultaneously,considering the latest active user had the greatest impacts on the inactive user in the process of being activated,the conception of influence weight delaying with time was introduced,to dynamic the process of influence weight accumulation,and to reduce the total number of active user that had been activated repeatedly.(3)In experiment,the two proposed algorithms were compared with several other algorithms respectively.The results verified the two-stage opinion leader mining algorithm can identify the opinion leader accurately and reduce the computational complexity.And the linear threshold model based on user's role was also verified to quantify the different influence of user role on information diffusion process availably,as well as decrease the number of active users that were activated many times.