Weibo Theme Mining Based On Dynamic Labeled-LDA Model

With the constant increasing of the users of microblog. The influence of Weibo is continually expanding. Because of the unique text feature of microblog class dynamic short text. The theme of blog are greatly affected by the time factor.Labeled-LDA model is a modification of the conventional LDA model. It is add an additional layer of category labels on the basis of the original LDA. It has an advantage of a direct distribution model categories more than the traditional LDA model. And a better solution to the traditional LDA model is applied to text classification problems will occur when classes are not on their own be forced to generate an implicit theme of the assignment.The purpose of this paper:since the short text of micro-blog network information is timelines, dynamic and theme complexity and diversity, this paper constructs a dynamic Labeled-LDA model, first, after reprocessing the selected microblog text, select the text feature. The dynamic improved VSM feature selection method is significantly better than the traditional TF-IDF algorithm.Secondly,this paper constructs a dynamic Labeled-LDA model, combined label with a dynamic state.The method consists of three parts:firstly, use Labeled-LDA topic model for modeling and feature selection, determine the number of the topic and hidden topics-text matrix; secondly, the the micro-blog text is divided into time text set; thirdly, improveme theLabeled-LDA model dynamically, and finally proposed a new topic search model for micro-blog text. Experiments conclusion:Time has a great influence on the microblog, Improved dynamic model can be seen more clearly changes with the theme of the time factor. The dynamic labeled-LDA model proposed in this paper for dynamic micro-blog class text theme mining has optimization effect.