Topic Detection And Tracking Based On Semantic Framework

With the rapid development of the network, how to guide the network public opinion effectively is becoming an important research direction of social security management. Now network public opinion analysis technology has developed into a research focus at home and abroad. And the technology has made some achievements in some fields, such as Topic Detection, Topic Tracking, Automatic Summarization and Public Opinion Warning System. Based on existing research on Topic Detection and Topic Tracking, this paper has made depth analysis in these two fields to provide effective support for the monitoring of network public opinion. What has done in the paper is mainly about the following two aspects.In view of the standard features of the news’ structure, the paper has proposed a new vector space model based on news’ semantic structure. The news documents are expressed as the four-dimensional vector space model including time, the person’s name or organization name, location and the news trunk. The paper calculates the eigenvalues according to the structure information and uses incremental clustering algorithm to detect topic. In the clustering process, this paper calculates the various dimensions of similarity and then weights the sum of these similarities as the final similarity. The heat formula has been proposed to achieve the topic from sets of events. Ultimately through the experimental comparison, it proves the fact that the program has improved the accuracy of the Topic Detection effectively.Based on the law of gradual evolution of topic centers on the timeline, the paper has proposed double centroid topic center representation model. The paper puts forward the concept of different degrees to find the topic center critical document and uses document distinct characteristics to distinguish the topic center effectively. Using incremental clustering algorithm to do secondary clustering for Topic Tracking, this method has improved the efficiency of Topic Tracking. Through the experiments, the paper finds that the program demonstrates the evolution of the topic centers.At last this paper details the development process and technologies used in the developing of popular references module of the real-time monitoring system on public opinion of microblog.