【作者】 王占一； 徐蔚然； 刘东鑫； 郭军；

【Author】 Zhanyi Wang,Weiran Xu,Dongxin Liu,Jun Guo PRIS Lab,Beijing University of Posts and Telecommunications,Beijing 10086

【机构】 北京邮电大学；

【摘要】 垃圾短信过滤是一个文本分类问题。如何设计分类器,可以在训练样本较少的情况下达到可观的精度是一个重要的问题。本文对传统分类器的结构加以改进,设计了一种基于潜在中间层的两级分类器,每级用贝叶斯方法实现。进一步地,将朴素贝叶斯分类器与之加权结合。实验结果表明,两级分类器大幅度提升了分类错误率的收敛速度。组合分类器在此基础上提高了训练样本较多时的精度,集成了两者的优点。更多还原

【Abstract】 Spam message filtering is an issue of text categorization.How to design a classifier that can reach a high precision when the training samples are not enough is a crucial problem.In this paper,the structure of traditional classifier is improved.A two-stage classifier based latent topics was implemented by Bayes method in every stage.Further more,it combined Na(i|¨)ve Bayes classifier through weighing.Experiments showed that the two-stage classifier significantly increased the converge rate of the error rate.Besides that,combined classifier raised the precision when the training samples were numerous and integrated the advantages of them.更多还原