Approach For Topic Tracking Based On Semantic And Hyperlink

As a new direction of research on natural language processing, Topic Detection and Tracking aims at developing technologies for event-based information organizing such as detecting stories on novel topic and tracking stories on known topics. Topic Detection and Tracking(TDT) can organize the distributed information so that people can grasp all the details about events and the relations between events.Topic tracking is a subtask of TDT. It aims at finding related stories on a certain topic that is identified using several sample strories. This paper focuses on the study of topic tracking task, makes the research on the characteristic of event news and advanced topic tracking system of other research institutions. Semantic and hyperlink analysis were introduced to the topic tracking method in this paper. The use of hyperlink analysis makes the method more usefull for the webpages, and the use of semantic analysis made the topic/event modle more intuitive and more specific for news.The experiments shows that the approach based on semantic and hyperlink improves the quality of topic tracking.Webpages are different from the traditional text documents. Some webpages have many photos and hyperlinks and only a few words, this means that the traditional topic tracking method based on content analysis is difficult to do well. Therefore, this paper studies the application of hyperlink analysis in topic tracking, and presents an approach combining the hyperlink analysis with content computing.The VSM oftern has high dimensions, in which important features often be inundated, and is not visual enough for the events, this paper presents an approach which used a semantic frame as event representation for topic tracking.Finally, the thinking of topic tracking is applied to theses, the application of topic tracking should not be just limited to news reports. This paper discusses the method of term weight calculation that exploited the featrues of theses, uses an expansion method of topic model based on synonymous. The algorithm of tracking based on citations is improved from KNN. Experiment proves that the application was reasonable.