Research On Sentiment Tendency Of Online Public Opinions

With the rapid development of Internet technology, the network media on behalfof the "new media" has increasingly become the main access to information andcommunication channels. Network has become a sharp "two-edged sword", providinga convenient way up to the higher authorities, but also on the country's political andcultural security pose a serious threat. Government to control irregular network ofemergency public opinion, based on the analysis of public opinion, the scientificdevelopment of appropriate early warning, reporting and feedback mechanisms, willplay a positive role in pacifying the public and restoring social order and stability.Traditional network public opinion research, mainly use the methods ofinvestigating related comments scale based on topic cluster, often lack of attention toemotional factors of the internet users'comments. Usually stay the extent of a hot topicand the hot topic found, no further distinction according to characteristics of emotionalbias. This article attempts to do in-depth study text analysis technology into thenetwork orientation of public opinion, according to public opinion corpuscharacteristics of the network, through machine learning methods obtain the emotionaltendency analysis model of public opinion on the network, then the emotionaltendency of public opinion on the network to do the correct prediction.This paper based on the correlation analysis theory and analysis process of publicopinion on the network, established a real estate public opinion corpus. Throughsummarizing and organizing the domestic and international related research literaturesand results, chose three model approaches for the text sentiment tendency of networkpublic opinion, and then through comparative analysis in experiment method to get thebest method for network public opinion, that Support Vector Machines. As the machinelearning involved in the training set and test set, in the sentiment analysis of networkpublic opinion explores the relationship between the number of training set and theclassification accuracy; the relationship between the number of negative comment andthe classification accuracy. This be good for improving the accuracy of sentimentclassification and reducing the classification costs. At last, applied the conclusion intothe case, analyzed the sentiment tendency of the text in the May 2010, and providedthe corresponding recommendation.In this paper, do in-depth research on the sentiment tendency of online reviews topublic opinion, on the one hand to enrich the theory field and content of networkpublic opinion on the network, which has some theoretical significance and value; onthe other hand, has the practical reference value for the development of monitoring on the network public opinion.