基于用户反馈混合型垃圾邮件过滤方法

《基于用户反馈混合型垃圾邮件过滤方法》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、基于用户反馈混合型垃圾邮件过滤方法　　文章编号:10019081(2013)07186105doi:10.11772/j.issn.10019081.2013.07.1861摘要：针对目前垃圾邮件过滤技术仅依赖单一邮件特征实施邮件分类、对邮件特征变化的适应性较差等局限，提出一种基于用户反馈的混合型垃圾邮件过滤方法。以用户社会网络关系为基础，借助用户反馈机制分别实现对基于内容与基于身份标识的邮件分类知识的动态更新；在此基础上采用贝叶斯模型，实现邮件的内容特征与发件人身份标识特征在邮件分类中的有机结合。实验结果表明，与传统的过滤方法比较，所提方法在邮件特征

2、动态变化的环境下能够获得更好的邮件分类效果，邮件分类的总体召回率、查准率、精确率均能达到90%以上。所提方法能够在保证邮件分类性能的同时，有效提高邮件分类对邮件特征变化的适应性，是已有垃圾邮件过滤技术的重要补充。关键词：垃圾邮件；基于内容的邮件过滤；基于身份标识的邮件过滤；邮件分类；用户反馈；贝叶斯模型中图分类号：TP181;TP393.098文献标志码:A英文标题4Hybridspamfilteringmethodbasedonusersfeedback英文作者名HUANGGuowei1*,XUYuwei2英文地址(1.ComputerColl

3、ege,ShenzhenInstituteofInformationTechnology,ShenzhenGuangdong518172,China;2.CollegeofInformationTechnicalScience,NankaiUniversity,Tianjin300071,China英文摘要)Abstract：Severallimitationsexistinthecurrentspamfilteringmethods,suchastheyusuallyrelyononlyonetypeofEmailcharacteristictoreali

4、zetheEmailclassification,andhavepooradaptabilitytothedynamicchangesofEmailcharacteristics.Concerningtheselimitations,ahybridspamfilteringmethodbasedonusersfeedbackwasproposed.BasedontheSocialNetwork(SN)relationshipamongusers,thedynamicupdateoftheknowledgeforEmailclassificationwasach

5、ievedwiththehelpoftheusersfeedback4scheme.Furthermore,theBayesianmodelwasintroducedtointegratethecontentbasedandtheidentitybasedcharacteristicsofEmailintheclassification.ThesimulationresultsshowthattheproposedmethodoutperformsthetraditionalmethodintermsofEmailclassification,whenthe

6、Emailcharacteristicschangedynamically.Theoverallrecall,precisionandaccuracyratiosofthemethodcanachieve90%andabove.WhileguaranteeingtheperformanceofEmailclassification,theproposedmethodcanimprovetheadaptabilityofclassificationtothechangesofEmailcharacteristicseffectively.Therefore,the

7、proposedmethodcanactasausefulcomplementtothecurrentspamfilteringmethods.Severallimitationsexistinthecurrentspamfilteringmethods,suchastheyusuallyrelyononlyonetypeofEmailcharacteristictorealizetheEmailclassification,andhavepooradaptabilitytothedynamicchangesofEmailcharacteris