The Application Study Of Probability Graph Model In Topic Detection And Information Diffusion

With the rapid development of social networks, more and more people post and retweets information through the sina platform. So it is an important research that studying the social public opinion in a period of time, such that the people's emotional attitude on a topic, the change of topic event in social network, what kind role of people in information diffusion, and how to make the topic spreading more widely in social networks. these problems have more and more important social value and commercial significance. However, Microblog is short and has irregular ungrammatical forms, also has a strong user's attitude and a strong real-time interactive character. So as to the traditional topic detection model and information diffusion model cannot be accurately study the problem.Therefore, this paper uses the probability graph model method to study the problem and put forward two kinds of models to study, such as dynamic sentiment-topic detection probability graph model based on the LDA (DST)and specific topic users decision information diffusion (TUDIN)probability graph model.Dynamic sentiment-topic detection probability graph model is a Bayesian graph model, based on LDA, using many kinds of probability and statistics method, fusing uses' sentiment and social network's time characteristic. This model can use the observable variable, such as word variables, document variables, time variables to estimate the latent variables, such as the users' sentiment and the topic of document. Model ignored the lack of data, the influence of the short text, the link between words, and improved the accuracy of topic detection and sentiment tendency analysis. The experimental results show that DSTM model has the lower perplexity compared with some other existing models and prove the performance advantages of DST model.Particular topic user decision information diffusion probability graph model is a generative probability graph model, fusing users'social role in social network and information diffusion characteristics. The TUDIN model uses the Gibbs sampling method to estimate and study the model on the historical data of information diffusion. The simulation experiment shows that TUDIN model greatly improves the precision of information diffusion and proves the model performance, compared with the existing methods of information diffusion, such as IC model, support vector machine (SVM) model.