一种基于容错粗糙集聚类方法的话题检测和跟踪框架.pdfVIP

一种基于容错粗糙集聚类方法的话题检测与跟踪框架宰 毋非1封化民1，2申晓晔1 1．西安电子科技大学西安710071 2．北京电子科技学院北京100070 E-maih wuf@besti．cn 摘要：文档聚类是实现话题检测与跟踪的重要方法。传统的硬聚类算法在处理边界数据时存在很大缺陷，影响话 题检测的效果。针对这一问题，本文提出一种改进的基于容错粗糙集的聚类方法，利用某些文档特征项协同出现， 构造特征项容错关系，扩充特征项的描述能力，最后用特征项容错类描述文档之间的相似关系，实现了话题检测 与跟踪的过程。实验证明，该方法能够有效的处理边界数据，提高文本聚类的效果。 关键词：话题检测与跟踪，容错粗糙集，聚类 A i Basedon Frameworkof i CDetecti onandTrack Top ng ToI erance SetCI usteri Method Rough ng l WUFei Hua-min 1．FENG u，SHEN Xiaoye 1．SchoolofTelecommunication Engineering,XidianUniversity,Xi’all710071，China and ElectronicScience 100070,China 2．Be[iilag Technologylnsfitmion,Beijing E-mail=mff@besti．cn Abstract：Textisall for Detectionand hard doesnot clustering clusteringimportanttechnologyTopic Tracking．Traditional the take betweenclustersinto leadstoinaccurate results．Inthis overlapconcepls aceoun‘which clustering paper．绷 ToleranceSetis featuresof K-means for Detectionand basedon improved algorithmTopic Tracking Roughproposed．The termsco-oceutl-encearcused the classes thesimilaritiesbetweendocumentsare todescribetoleranceofterms．And given thetermstoleranceclasses。The resultsonmid-scaledatasets