Propagation Effect Analyzing And Predicting On Microblog

As a new social media, Microblog can be used to access, share and propagate information. The process of diffusion has fission characteristic, messages diffusion by publishing, commenting and retweeting. Therefore, microblog has faster diffusion speed and wide diffusion range; it makes traditional medias be cast into the shade. Research on propagation effect of microblog, we can exploit the results to analyze and predict the influence of news.In microblogging network, retweet behavior is the main way to spread messages. So exploring the user’s retweet behavior is benefited to research retweet scale comprehensive and thoroughly. Predicting the scale of tweets, we can find a hot topic in the message diffusion process and accurately predict the development trend of a microblog. Furthermore, the government can effectively intervene and control the spread of public opinion, to control information reasonably. Our work mainly focuses on two aspects: on the one hand, through the analysis of the factors affecting user’s retweet behavior, and then predict whether a user will forward a microblog. On the other hand, proposes a final retweet scale prediction model based on the influence factors, using retweet scale as a measure of the propagation effect.When we study whether a user will retweet message m or not, firstly, this thesis makes a brief description of the problem to start from the research questions, and then makes a clear definition of the user’s retweet behavior and unretweet behavior. Secondly, a number of verification for the effect of extracting factors on user behavior, and then add some basic factors to constructed a comprehensive indicator system of influence factors of retweet. These factors divided into four categories as user profiles, microblog text features, interactive attributes and local structures for predicting user’s retweet behavior. Finally, using these multifeatures, we utilize supervised classifiers to predict retweet behavior on Sina Weibo dataset. The results show that the features we selected combined with the Logistic Regression model for predicting user’s retweet behavior more accurately.About propagation effect of microblog, most of the current studies consider the scale prediction as a classification problem. In this paper, we can predict the specific microblog into one of the four classes, using a classification model. And then, we propose regression models to predict retweet scale for each type of microblog. With the main factors that affect user behavior and other scholars’ research, not only static features include user profiles and microblog features be considered, but also dynamic attributes of microblog after a period of time be considered. Because there are many factors that affect the retweet number, we choice the regression prediction model based on ridge regression. The experiment proved that the model is feasible and effective.