Dynamic Community Analysis And Popularity Degree Prediction For Online Social Network

Microblog is one of the most popular kinds’of social media services nowadays. Compared with traditional media, microblog has many unique characteristics such as a large amount of users, fast sprread speed, brief messages and so on. The scale of microblogging network increases tremendously in recent years.Therefore, the microblogging network become a fertile soil for information diffusion and has received widespread attention. Detection and understand mechanisms of information diffusion and dynamics community structure evolution mechanism of microblogging network.It is benefit for grasp the law of dissemination of information, prediction transmission path, discovery networking groups events, control network warning public opinion and so onThis paper focus on information network formed by microblogging forwarding. We do study on n information network from multiaspect,such as information diffusion,popularity and dynamics community structure evolution.Firstly, we analysis a fast greedy algorithm for Imodularity optimization designed by Blondel et al and propose an optimizatiion algorithm to detect communities in large networks. By tests we see that different node order brings different performance and different community structure. We find some nodes swing in different communities that influence the performance. So We introduce a new concept OV that shows the strength of connection between nodes, we design some strategies on the sweeping order of node to reduce the computing cost made by repetition swing。 Experiments on synthetic datasets and real datasets are made, which show our improved strategies can improve the performance and correctness.Secondly, we explored the community network structure of microblog information dissemination network from the macro.we constructed a new information diffusion network according to microblogging forwarding path, different from the traditional networks. We do dynamic community discovery on networks snapshots which is divided by one month intervals to analyze the evolution of the community.Thirdly, we do empirical analysis on Sina Weibo Datasets. We studied the popularity of tweets in microblogging network and introduce a novel concept "popularity degree". Through the empirical analysis of different popularity degree, we find the retweeting information of a tweet at an earlier time can help predict its final popularity.Finally,we propose a model based on SVM with the retweeting information within one hour. Experimental results show our model has better ability of prediction. In future work, we will continue to further study and improve its ability to predict the highly popular microblog.