Analysis Of Weibo Sentiment Polarity Based On Multi-strategy

Weibo is a social networking platform of strong timeliness and informality. The earliest weibo in china is appeared in 2007 and has experienced rapid development. Weibo contains a large number of information about various fields. Through the sentiment analysis of weibo,we can dig up a lot of valuable information such as user’s opinion of goods,social events or policies which are formulated by government.Thus,the sentiment analysis of weibo has become very realistic and far-reaching significance.In this paper we study sentiment polarity classification of weibo and put forward a multi-strategy weibo sentiment analysis way,the main contents of the research include the following points:The first part: We study unique linguistic features of weibo for the problem of weibo sentiment classification, we use the commonly weibo emotional polarity classification methods based on weibo facial expressions symbols, emotional dictionary, improved support vector machine to estimate sentiment polarity of weibo.The second part: To extend and improve the basic sentiment lexicon used in experiments. We put the commonly used expression symbols, basic emotional lexicon and network convergence sentiment words together to form a new weibo sentiment dictionary.Then we build a text word library which include new weibo sentiment dictionary,negative word dictionary and the degree of word dictionary etc.The third part:Study and improve the weibo sentiment classification method based on SVM, select basic sentiment words, network sentiment words etc as sentiment polarity classification feature of weibo. We get the optimal values of SVM kernel function parameters through many experiments on different data to improve result of view and polarity classifier.The fourth part:Study the influence of training set and values of kernel function parameters on the result of sentiment polarity classification experiments.The experiments results show that the best accurate rate of classification experiments is 83.1%,the result is better than that of experiments based on sentiment dictionary and weibo facial expressions symbols, the effect of experiment is considerable.