Topic and Sentiment Unification Maximum Entropy Model for Online Review Analysis

Abstract

Opinion mining is an important research topic in data mining. Many current methods are coarse-grained, which are practically problemic due to insufficient feedback information and limited reference values. To address these problems, a novel topic and sentiment unification maximum entropy LDA model is proposed in this paper for fine-grained opinion mining of online reviews. In this model, a maximum entropy component is first added to the traditional LDA model to distinguish background words, aspect words and opinion words and further realize both the local and global extraction of these words. A sentiment layer is then inserted between a topic layer and a word layer to extend the proposed model to four layers. Sentiment polarity analysis is done based on the extraction of aspect words and opinion words to simultaneously acquire the sentiment polarity of the whole review and each topic, which leads to, fine-grained topic-sentiment abstract. Experimental results demonstrate the validity of the proposed model and theory.