The Study Of Sentiment Classification On Chinese Microblog

Microblog as a new information publishing and social networking platforms, sinceits birth, the scale of its users soared in these few years. After registration ofmicroblog, users make friends through microblog, concern friends and celebrities,forward, reply, comment news.Microblog has become an important way to obtaininformation, interact with other people and express user’s own feelings and opinionsby majority of Internet users.It is to penetrate and affect from all aspects of people’slives.With the growing of commercial value, media value, social value and influence ofmicroblog, it attracts more and more scholars to study it. Sentiment analysis is one ofthe important issues, and it is to identify for emotion polarity in microblog, whichbelongs to the positive, negative or neutral. Through analyzing microblog forsentiment analysis, the brand promotion, microblog marketing, public opinionsupervision and so on can be realized.The paper takes hot topics in Chinese microblog as research objects, studyssentiment classfication characteristics and methods of Chinese microblog. In theclassification method, we verified the method which was based on expression andemotion dictionary rules, and the approach based on SVM classification. Weconsider two kinds of programs, according to network language features and featureextraction of microblog. One is by building the relevant emotional vocabularyresource dictionary and expression, we select keywords, expression, emotionalwords, phrases, punctuation, and part of speech of emotions as a semanticcharacteristics.The other one is considering feature extraction of term and emotionalcategories dependency which based on Unigram and CHI statistical.The experiment is over NLP&CC2012evaluation data.The results show that theSVM classification based on microblog semantic feature, the classification effectdepends largely on emotional dictionary, expression and other resources.The effortsof SVM classification based on Unigram and CHI statistical, is slightly better thanthose based on microblog semantic features, and the accuracy is improved by2.6%.