Analysis of document sentiment based on probability graph model and LSTM

• 1、State Key Laboratory of Networking and Switching Technology, Beijing University of Post and Telecommunications

Abstract：Document sentiment analysis plays an important role in many fields such as economy, entertainment and politics. For the document sentiment analysis, the contextual relationship is far away and the emotional transition is sharp. This paper first uses the improved LDA topic model to extract the emotional theme of document and convert long text into topic sentence. Then use the LSTM neural network structure to analyze the emotions of the document topics. Finally get the emotional polarity of the document. The experimental data was tested using the Douban Movie Review Data Set. Emotional polarity includes positive, negative and neutral emotions. The experimental results show that the proposed method is better performence to the traditional sentiment analysis method, and the classification precision rate is 47.64%.

Keywords： Document Sentiment Analysis LDA LSTM