Research On Key Technologies Of Topic Detection And Tracking

With the proliferation of computers and the rapid development of internet, information is explosive growth, there are more and more attention to topic detection and tracking. Topic detection and tracking technology mainly make lots of unordered information be obtained, cleared up, organized timely and accurately to become the form easy to see about and search, so that users can make the analysis of mutual evolution and development tendency among the topics, which provides higher level service for the users. The technology is also a proposed applied research in response to information overload.In this paper, research the relevant content of the topic detection and tracking. First of all, a novel approach is used based on the cascaded model for the topic detection to effectively distinguish similar topic problem. This approach fully considers the importance of the location, time and other entity information of the report in the topic detection, By constructing a cascaded model, at the bottom level of the system, based-content similarity was used to realize topic detection task. At the top level of the system, three similarity results were effectively combined to realize topic detection task. Experimental results show that the performance of this novel algorithm is superior to the traditional text similarity algorithm.Secondly, the classification strategy is used in the topic detection. Support vector machine as a classification strategy, this paper summed up the advantages and several methods which based on multi-class SVM, a novel approach is proposed, which is mainly based on SVM and rules to realize topic detection task. Experimental results show that this novel algorithm is effective to improve the system stability and recall.Finally, an algorithm of topic tracking based on topic renewal is used to avoid the sparse of the topic priori knowledge and the topic drift. This method improves the algorithm of topic renewal, using secondary threshold set to realize adaptive topic tracking. Experimental results show that this novel approach is effective to improve the efficiency and the system stability of the topic tracking.