摘要： 针对互联网上论坛和新闻网站发布的海量自然语言文本，该文设计一个话题识别与跟踪系统，将海量的数据分类整理并聚合形成各个话题。该系统的核心采用SVM方法进行文本分类，基于知识库和网络流算法实现话题的聚合，测试结果表明，文章分类的正确率达到92%，聚类的正确率达到88%，具有较高的应用价值。

关键词: 话题识别与跟踪, 信息检索, 支持向量机, 分类, 聚类

Abstract: This paper designs and implements a Topic Detection and Tracking(TDT) system to process the huge number of natural language text on Web. It classifies the text into several categories, performs clustering in each category to get the topic. The system can detect the hot topics in real-time and track some topics selected by user. The accuracy of text classification is 92%, and the accuracy of clustering is 88%. Experiment shows the feasibility of the TDT system.

Key words: Topic Detection and Tracking(TDT), information retrieval, SVM, classification, clustering

中图分类号: 

• TP393