Abstract:

Many current methods of opinion mining are coarsegrained, which are practically problematic due to insufficient feedback information. To address these problems, we propose a novel topic and sentiment joint maximum entropy LDA model in this paper for finegrained opinion mining. Considering semantic and location information of words, a maximum entropy component is first added to the traditional LDA model to distinguish background words, aspect words and opinion words. Both the local extraction and global extraction of these words are further realized. Secondly, a sentiment layer is inserted between a topic layer and a word layer to perform finegrained opinion mining on word or phrase level. Transition variable is introduced to deal with sentiment dependency. The sentiment polarity of the whole review and each topic are simultaneously acquired. Experimental results demonstrate the validity of the proposed model and theory.

Key words: LDA model;fine-grained opinion mining;maximum entropy;sentiment dependency