Research On Multidimensional Microblog User Influence Analysis Techniques

Microblog is a significant application based on user relationship to get, share and broadcast information, such as Twitter and Weibo. It has characteristics as strong real-time performance, strong interactivity, fast spreading speed, high updating frequency, wide range of influence, etc. In this thesis, we take Weibo as research platform and focus on the problem of user influence analysis from dimensions: the friendship network, user comment behavior and contents user posted. 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) Designed and implemented a user influence analysis algorithm ANR, based on user comment behavior. The algorithm uses the user comment behavior to build an activity network AN, and compared with user friendship network model, the AN model can better depict user behavior in information dissemination in Weibo. Based on the random walk model, using the weight value of nodes and edges in AN model, the ANR gives different jump probability between nodes, which better describes the effect of node connection strength, activity and other properties in user influence analyzing, improves the precision of the results of the analysis. Experiments show that compared with the user friendship network based algorithm IND and PageRank, the top 20 analysis results of ANR algorithm have a better average posts number and comments number, increased an average of 2.39 times and 1.66 times with IND and PageRank.(2) Designed and implemented a user influence analysis algorithm KsRN. The algorithm constructs a network model RN, in which user comment behavior is used to update the user friendship network in the fact that the friendship network cannot reflect the user activity in information diffusion. Then RN is transformed to URN, from a directed network to an undirected one, for the support that node centrality methods like k-shell decomposition can be used for user influence analyzing in Weibo. KsRN has a biggest correlation ratio of 82.7% with the actual measurement INF, and has an average ration improvement of 33.26% and 37.68 compared with friendship network based user influence analysis algorithm IND and PageRank.(3) Designed and implemented a multidimensional user influence analysis algorithm MDUIR, which combines the effect of user friendship, user comment behavior and contents user posted. The algorithm put forward a SIM method to measure the significant index of user behavior using the created time, content quality and correlation with the topic. SIM is used in the construction of the IN network model, which combines the user friendship network and user comment behavior, and figures out the issue that network models like user friendship network, the AN network and the RN network cannot reflect the effect of contents in user activity and time damping in user influence analysis. Further, MDUIR applies the random walk in the IN network model to analyze the user influence. Experiments show that MDUIR makes full use of the effect of contents user posted and time damping in user influence analyzing process, and the rank results of MDUIR are more accuracy and rational than that of IND, Page Rank, ANR, and Ks RN.From the conclusions of this thesis, conclusion can be include that user influence analyzing methods that considers user behavior and contents are better than those only using the user friendship network.