Recognition Of Chinese Microblog Sentiment Polarity And Extraction Of Opinion Target

With the rapid development of Internet technology, Microblog has become the majority social platform to publish news, express views and share experiences. More and more internet users are interested in expressing their views and opinions such as government policy, a hot topic etc. by updating Microblog. And Microblog included various domains, just as social, economic, cultural and so on. Analyzing Microblog text with sentiment can provide the reference for potential customers such as Government departments, enterprises and individuals. Also Microblog sentiment analysis has become a hotspot for natural language processing. Oversea Microblog sentiment analysis has entered into applications. For example, 2009 Germany election, Tumasjan, analyzed the 100,000 related Twitter-texts, public opinion accurately reflected the election results. Chinese Microblog sentiment analysis is still in the initial stage. Based on the analysis of Chinese Microblog sentiment classification and evaluation object recognition, in this paper, the author uses the maximum entropy model to recognize sentiment polarity of Chinese Microblog text and uses the conditional random fields to extract evaluation object. This study includes the following aspects:(1) Building dictionary of Chinese Microblog sentiment words and dictionary of Internet terms. These dictionaries are the key to recognize sentiment polarity of Chinese Microblog text and extract evaluation object.(2) Using rule-based and machine learning methods to recognize Chinese Microblog text sentiment polarity. Based on sentiment words location and syntactic dependencies, the author extracts Microblog text features, and then uses these features to train the maximum entropy model and recognize Chinese Microblog text sentiment polarity.(3) Using machine learning methods to extract opinion target of Chinese Microblog text with sentiment. Based on words, part of speech and syntactical features, the author trains the conditional random field model to extract opinion target.With the Training and Testing Microblog corpus of COAE2013 task three & four, and compared with the result of COAE2013, the author's testing results have been significantly improved. This demonstrated the feasibility and effectiveness of methods.