Analysis Of User Influence Based On Topic Diffusion In Microblogging

Microblogging is a significant platform based on user relationship to get, share and broadcast information. In recent years, with the rapid development of microblogging, more and more people start to use it, which attracts many researchers from various fields to commit themselves into microblogging related work. However, it is a big challenge to study about it, as microblogging is a new type of online social network, with the large number of participants, the frequent updating of topics, the rapid diffusion of information, and the widespread impact. In this thesis, we mainly take Sina weibo as research platform and focus on the problem of user influence analysis in topic diffusion. This work first analysed the effection on user influence from information diffusion, individual attributes, and microblog theme, and then proposed a new algorithm of user influence analysis in microblogging. The conclusions in this thesis could be useful for the monitoring and tracking of public opinion in microblogging network.Three major contributions of the thesis are as follows:(1) We built a novel topic diffusion network model based on social relationship and user behaviors. Considering the two kind of information diffusion mechanism in microblogging, we firstly construct the information listenning network (ILN) and information forward network (IFN) with retweet behavior and social relationship between users, and then map IFN into ILN which can reflects the frequency of information exchange between nodes. Compared to ILN, this model could exactly describe the diffusion process of topics with user behaviors, which makes it easier to analyse user influence in specific topic.(2) We proposed a new method of user influence regression analysis named PBF, based on individual attribute features. The method firstly uses information diffusion characters to measure user influence, and then makes regression analysis of it with individual attribute features to find out the ones effecting user influence most, with which we then predict user influence. Compared to other index system, PBF method is more objective and accurate in user influence analysis overcoming the variety of individual attributes. The experimental results indicate that PBF method improves the accuracy of user influence analysis effectively. The highest correlation coefficient between results from PBF and mearsured influence can reach89.6%.(3) Based on topic diffusion network model mentioned above, we design and implement a user influence ranking algorithm named TS-InfluenceRank, whose basic idea is similar to PageRank. For the shortage of distributing transition probability equally in PageRank and LeaderRank, the algorithm calculates the transition probability using individual attributes and the correlation between user and topic. As for the convergence of algorithm and the problem of spider trap in PageRank and TwitterRank, this algorithm solves them by adding a ground node into the network. In addition, the join-in of ground node makes the probability of teleport self-adaptive to different nodes. With two real-world data sets from Sina weibo, we compare TS-InfluenceRank with LeaderRank, TwitterRank, PageRank, and In-degree. From the comparing results, we found that TS-InfluenceRank is more accurate and effective in user influence analysis within specific topic. Taking LeaderRank for example, the maximum improvement of ranking sequence correlation and influence proportion can be as high as13.9%and7.2%respectively.