LS-SVM:一种有效的新闻主题追踪方法

LS-SVM:effective algorithm of news topic tracking

新闻主题追踪是对主体所感兴趣的新闻主题的发展趋势进行动态追踪,其优势在于对所感兴趣的主题基于文本模型及理解的动态追踪,因此更多地涉及文本表示与语义理解。LS-SVM首先将文本利用LSI(隐含语义分析)进行分析,完成对文本基于语义的特征降维及文本表示;然后将隐含语义文本表示的结果输出给SVM进行主题追踪,从而实现从语义层次上的新闻主题追踪。实验结果表明,与传统的主题追踪相比较,该方法能够有效提高主题追踪的性能,减少追踪的错报率和漏报率。

Topic tracking is to track news trend which someone is interested in it. Its advantages lie in dynamic tracking based on text model and understanding, so it involves in more text express and semantic understanding. LS-SVM first analyzed text using LSI（latent semantic indexing）, which achieved semantic-based character reduction and text express, then combined SVM to complete semantic-based topic tracking. The result of experiment shows, compared to conventional methods, LS-SVM can improve performance of topic tracking effectively and reduce fault and fail rate of topic tracking.