Based on the analysis of the status of network public opinion, the features of short content and nearly real-time broadcasting velocity in this paper, it constructs a public opinion corpus on the content of micro-blog information, and proposes an approach to marking corpus on the basis of sentiment tendency from the semantic point of view; Furthermore, considering the characteristics of micro-blog, it calculates the sentiment intensity from three levels on words, sentences and documents respectively, which improves the efficiency of the public opinion characteristics analysis and supervision. So as to provide a better technical support for content auditing and public opinion monitoring for micro-blog platform.

On the features of short content and nearly real-time broadcasting velocity of micro-blog information, our lab constructed a public opinion corpus named MPO Corpus. Then, based on the analysis of the status of the network public opinion, it proposes an approach to calculate the sentiment intensity from three levels on words, sentences and documents respectively in this paper. Furthermore, on the basis of the MPO Corpus and HowNet Knowledge-base and sentiment analysis set, the feature words’ semantic information is brought into the traditional vector space model to represent micro-blog documents. At the same time, the documents are classified by the subjects and sentiment intensity. Therefore, the experiment result indicates that the proposed method improves the efficiency and accuracy of the micro-blog content classification，the public opinion characteristics analysis and supervision in this paper. Thus, it provides a better technical support for content auditing and public opinion monitoring for micro-blog platform.