Topic Model For Graph Mining Based On Hierarchical Dirichlet Process

Topic model has been proved to be very successful in text mining.Howerver,the classical topic models are based on the assumption of ”bag of words” which ignore the association between words.Besides,the number of hidden topics often needs to be predetermined.In practice,however,it is difficult and sometimes impractical to determine the appropriate number of topics.In this paper,a nonparametric Bayesian thematic model HDP-GTM based on Hierarchical Dirichlet Process is proposed.The Hierarchical Dirichlet Process makes the number of topics selected flexibly,which breaks the limitation that the number of topics need to be given in advance.Moreover,HDP-GTM releases the assumption of ”bag of words” and consider the graph structure of the data.Thus,more information in text data can be discovered and the text classification effect will be significantly improved.The validity and superiority of the HDP-GTM model proposed in this paper is further verified by experiments.