Chinese Sentiment Analysis Based On LSA And MEM

Current blogs, BBS, bars have become the important channels for people to express their opinions and emotion. Emotion information produced by these channels reflects people's attitude towards hot issues. Accordingly, to mining and analysis emotional information, we can better analysis hot issues, understand the user's interests, provide government and enterprises important information and decision support.The traditional information retrieval technology based on keywords, can not provide good support for this demand.Traditional information extraction and text classification techniques also were not related to deeply semantic understanding. Text Sentiment analysis is a kind of technology which analysis emotional information in text. It can find new hot events more quickly from mass information on net and grasp the orientation of public opinion. It improves subject tracking research and provides new ideas for text categorization, information extraction, and summarization. It has broad application space in enterprise intelligence analysis, government public opinion analysis, information security and automatic abstract.Sentiment analysis research involves in wide fields, this paper focuses on emotion word, sentence recognition and classification. This paper first analysis the research background and meaning of sentiment analysis at home and abroad.Secondly, this paper presents a word-level sentiment analysis based on the latent semantic analysis and maximum entropy model. Its main idea is to use latent semantic analysis method to calculate the semantic similarity of words, combined with several characteristic functions to establish maximum entropy emotional word recognition model. Besides, this paper puts forward Chinese emotional sentence classification method mixed various characteristics based on word-level sentiment analysis. Finally, the methods is tested and verified by using 935 experimental documents in COAE2008 testing set. The word sentiment classification gets an accuracy of 83.5% and a recall rate of 79.3%. The sentence sentiment classification gets an accuracy of 76.8% and a recall rate of 78.3%.