Research Of Sentiment Analysis For Chinese Micro Blog Based On Conditional Random Field

With the arrival of Web2.0, Internet is developing from simple Read mode to Write mode and Build together mode, users are developing from passively receiving information to actively creating Internet information. In the Web2.0era, Internet provides lots of open information communication platforms for users. With BBS, Blogs and Microblogs, users can share all kinds of opinions without the limit of time and space, they can not only achieve information they need, but also post their opinions. Among these platforms, Mieroblog is the newest, and is quickly accepted and recognized by people. Because it is convenient and fast, lost of users express their opinions on some hot topics and products, and these opinions always have strong personal emotion. It has great value to mine these massive emotion information, therefore sentiment analysis on Mieroblog information is pressing and important.Based on previous achievements on sentiment analysis, this paper analyzes and observes Chinese Mieroblog messages’structure features according to the open source Sina Weibo data sets. Then this paper proposes the disadvantages of existing research ideas. Different from the most existing works which split sentiment analysis into two independent tasks:Subjective classification and Polarity classification, in this paper, a DCRF model with a two-level structure is proposed, which can introduce sentiment poarity context information and dependence information between two tasks in previous works. As shown in the experiments, after introducing these two kinds of information, the precision is improved by four percent more than the state-of-the-art approach. This paper also compares the contribution of these two kinds of information, and it is found that the improvement mainly comes from the dependence information between two tasks.