【作者】 陈超兰；

【导师】 张自力；

【作者基本信息】 西南大学 ， 计算机应用技术， 2006， 硕士

【摘要】 电子邮件是互联网的最重要应用之一。它在给人们日常工作和生活带来很大便利的同时,也带来了一种令人讨厌的副产品——垃圾邮件。随着垃圾邮件越来越泛滥,人们在技术和法律方面都进行了不断的努力,垃圾邮件已经得到了一定的控制。 垃圾邮件过滤的主要技术包括白名单与黑名单技术、规则过滤、基于关键词匹配的内容扫描,以及基于内容的文本分类方法等。目前的垃圾邮件过滤系统如贝叶斯过滤系统等,从过滤效果来看,并不是十分理想,存在的主要问题是将非垃圾邮件判定为垃圾邮件的几率较高,使用户宁愿接收到垃圾邮件也不愿意使用邮件过滤系统。基于粗糙集的垃圾邮件过滤是一种基于规则的内容过滤方法,将粗糙集理论用于垃圾邮件过滤是一个新的研究方向,可以降低垃圾邮件错判率。 本文的选题正是基于上述背景,本文的工作主要包括以下几个方面: 1.给出了垃圾邮件的定义,探讨了垃圾邮件的危害; 2.分析并总结了当前主要的垃圾邮件过滤技术,对常用邮件分类方法的基本原理及分类准确率进行了介绍; 3.介绍了基于粗糙集的垃圾邮件过滤系统模型和工作流程,并在此基础上,对该模型进行了改进; 4.对基于粗糙集的垃圾邮件过滤系统中的特征选择问题进行了研究,提出了用Mitra’s+SFS算法来进行邮件特征选择,将过滤冗余特征和不相关特征相结合,提高了系统分类准确率; 5.利用基于Java的机器学习软件Weka,对所选出的特征子集进行分类实验,并对实验结果进行评估,验证了所提出特征选择方法的有效性。更多还原

【Abstract】 Electronic mail (e-mail) is one of the most popular services of the Internet. E-mail has brought us great convenience in our daily work and life. At the same time,It has brought us an annoying byproduct-Spam(also referred to as "junk mail").Because we have devoted ourselves into the task of anti-spam by the way of technologies and law, the spam has been in our control in some degree.Nowadays, anti-spam measures commonly include black or white list technology, manual rules and keyword based content filtering. Many email filtering systems such as Bayes filtering system, are not very ideal in spam filtering effect. The main problem in these filtering systems is the possibility of discriminating non-spam to spam is high, which causes users would not use email filtering system. Rough set based spam filtering is one of the rule based content filtering methods. Applying rough set to spam filtering domain is a new research task, it can reduce error rate of classifying a non-spam to spam.With the above observations in mind, the work of this thesis is as follows:1. Discussing what is spam and its harm;2. Presenting typical anti-spam techniques and discussing the fundamental principles of email classification and their classification accuracy;3. Present rough set based spam filtering model and its work flow, and on this base, improve the system model.4. Research on feature selection problem of rough set based spam filtering system , present a new feature selection method, which combines both Mitra’s and Sequential Forward Selection and improve system classification accuracy.5. Taking experiments for several feature selection algorithm with Weka (a machine learning software based on Java), evaluate the experiment results and validate the proposed algorithm.更多还原