Research On Sentiment Analysis For Chinese Microblog

Microblog is becoming a most popular internet application. According to the statis-tics, more than100million tweets publiched in everyday. These tweets not only conveythe description of facts, but also contain the emotional states of massive microblog users.And these emotional informations may be help for user to decide whether buy a product,provide very important reference value for companies to make market strategy, and evenmake massive data available for government to monitoring public opinion.In light of this, we proposed a sentiment analysis method based on a combination ofsyntactic dependencies and text classification techniques for Chinese tweets. The methodadopts the syntactic dependencies to perform sentiment analysis, at the same time, com-putes a confidence for every tweet. Choosen tweets which confidence above a certainthreshold as training samples, train a two-step sentiment classifier by using the contentfeatures and media features of tweets. Finally, classify the sentiment orientation of tweetsagain. In addiation, we also proposed a method that serves common emoticons as the sen-timent class labels of tweets and implements an incremental learning method to tackle theproblem of real-time sentiment analysis.Experimental results show that the proposed method dramatically improves the pre-cision and the recall by6%and3%repectively compared to the method that only based onsyntactic dependencies. And the performance of our two feature sets are also better thanunigram features, the precision and the recall both are88%in term of subjective classifier,and they are72.1%and71.5%for sentiment classifier. Apart from this, the media featuresare good for trackling the problem of real-time sentiment analysis.