Study Of Text Evolution Analysis And Prediction Based On Topic Model

With the development of the Internet, social network has a rapid improvement. Evolutionary analysis of mass data generated by the users in social network has become a hot spot in natural language processing. In social network, traditional topic detection method is difficult to guarantee its accuracy and efficiency. Therefore, topic model method is proposed. Topic model method thinks that the topic is the distribution of a group of semantically related words and the probability of them.The characteristics of semantic information in large-scale corpus can been simulated well, so it has been widely applied.Based on the existing topic acquisition and evolution, the following meaningful innovation is been proposed in this paper. First, a topic evolution method based on the latent Dirichlet allocation model. This method on the foundation of the existing evolution method, adds to the method of topic distillation, and in the topic evolution presents a method to judge the theme evolution state. Then,combined with the Markov chain prediction model, propose a method of topic evolution trend forecasting based on topic evolution method. Last but not least, try to use the hierarchical Dirichlet process method to replace the latent Dirichlet allocation model, to achieve the automatic verification number in the process of topic evolution.Sets of experiments in the NIPS papers show that evolution is proposed in this paper can accurately obtain evolution trend of the topic in content and intensity, combined with the Markov chain prediction method can effectively predict the evolution trend in the next several time windows,using hierarchical Dirichlet method to analysis and predict the topic evolution is better than that of using latent Dirichlet allocation model.