Context Based Sentiment Analysis On Chinese Micro-blog Text

With the popularity of the Internet and mobile devices and wireless networks, social network services, such as the micro-blog, have become more and more popular. The rapid growth of users and a large number of comments,making micro-blog text research work is also increasingly concerned by scholars,micro-blog emotional analysis task is one of the important research.People express their opinions and emotions on the micro blog,which is often spoken colloquial serious and strong subjective emotional color.Micro-blog can reflect the individual most real instant speech, full of high commercial and research value, and micro-blog sentiment analysis is the process which obtain valuable information through a series of technical means to judge the micro-blog text emotional polarity.At present, foreign micro-blog emotional analysis related research is relatively mature, the domestic is still in its infancy. More and more people express their opinions and emotions on the micro blog, so that the sentiment polarity judgment of micro-bloggers becomes more and more important. Existing traditional algorithms for the analysis of Chinese micro-blog sentiment do not take into account the context of the micro-blog text, so can not identify the sentiment of micro-blog very accurate. Therefore, in order to utilize the context of the micro-blog text, this paper presents a context based sentiment analysis model on Chinese micro-blog text. This model take into account the context of the micro-blog text by using the SVMhmm algorithm, in which the micro-blog sentiment analysis problem is regarded as a label sequence learning task. Experimental results show that, micro-blog sentiment analysis model results based on SVMhmm algorithm can make better analysis of the micro-blog polarity than micro-blog sentiment analysis model based on Naive Bayes or SVM algorithm. Among them, the SVMhmm with SGNS text representation method do best in performance. The method accuracy and F1 value are both 0.69 on the whole micro-blog datasets, while its accuracy rate is 0.71, F1 value can reach 0.70 on the rich context micro-blog datasets.