Web public sentiment reflects the people's attitude to society and politics, it's important to identify the web sensitive information correctly to build a harmonious network. Those existing monitoring systems relied on large common category libraries to do accurate identification, yet faced with the new and unknown events which is not recorded in the category libraries did not do in-depth research. In view of this situation, this article proposes an adaptive method combination of content identification and acts recognition to achieve accurate and effective identification of web sensitive information.