基于用户关系行为的垃圾邮件判别机制的研究与实现

发布时间：2018-04-04 02:08

本文选题：垃圾邮件　切入点：用户关系行为　出处：《东北大学》2014年硕士论文

【摘要】：作为互联网最基础的应用之一,电子邮件的广泛使用使其占据互联网数据总量的三分之一左右。随着应用的增多,垃圾邮件数量的增长也非常迅速,网民每周收到的垃圾邮件占总邮件量的三分之二还多。近几年来,为应对垃圾邮件愈禁愈多的情况,众多研究人员提出了多种有效判别垃圾邮件的解决方案,但随着实际情况的变化,垃圾邮件发送形式和方法不断演进,各类技术总是无法长期发挥作用。本文总结了现有垃圾邮件判别技术,重点分析了邮件用户关系行为,在现有基于行为解析的垃圾邮件判别技术的基础上,归纳出邮件用户关系行为网络社区概念。根据邮件用户之间的关系,建立邮件用户关系行为网络,利用复杂网络社区划分算法划分用户关系社区,然后把这些用户关系社区应用于实际的垃圾邮件判别。本文在研究现有行为识别技术的过程中,把邮件通信行为网络看成是人际关系网络范畴,该判别机制的关键在于用户关系网络的社区划分。为此,引入快速Newman算法(简称FN算法)、CPM (Clique Percolation Method)算法和基于节点联系算法,建立对比社区划分机制,以获得最优划分。将建立的用户关系社区划分机制的具体模型应用到反垃圾邮件领域中,提高了垃圾邮件的识别精度和处理速度。为了提升判别机制的准确度,对通过用户关系行为社区识别出的可疑邮件采取进一步的判别,引入了分类回归树算法(CART),根据前期邮件头数据处理获得邮件行为特征数据,对照正常邮件行为进行多重判别。经过系统实现验证,结合测试数据的反馈结果证明,基于用户关系行为的垃圾邮件判别机制的设计是合理有效的,不仅减轻了邮件服务器的过滤负载,而且具有较高的准确率和很好的召回率。
[Abstract]:As one of the most basic applications of the Internet, email accounts for about 1/3 of the Internet data.With the increasing number of applications, the number of spam is also growing rapidly, and the amount of spam received by netizens accounts for more than 2/3 of the total number of emails each week.In recent years, in order to deal with the increasing number of spam, many researchers have put forward a variety of effective spam discrimination solutions, but with the change of the actual situation, the form and method of spam sending has been evolving.Technologies of all kinds can never work for a long time.This paper summarizes the existing spam discrimination technology, analyzes the email user relationship behavior, and summarizes the concept of email user relationship behavior network community on the basis of the existing spam discrimination technology based on behavior analysis.According to the relationship between email users, the behavior network of email user relationship is established, and the user relationship community is divided into user relationship communities by using complex network community partitioning algorithm, and then these user relationship communities are applied to the actual spam discrimination.In the course of studying the existing behavior recognition technology, this paper regards the mail communication behavior network as the category of interpersonal relationship network, and the key of this discriminant mechanism lies in the community division of the user relationship network.In order to obtain the optimal partition, the fast Newman algorithm and the node connection based algorithm are introduced in this paper.The specific model of user relationship community partitioning mechanism is applied to the anti-spam field, which improves the identification accuracy and processing speed of spam.In order to improve the accuracy of the discriminant mechanism, the classification and regression tree algorithm is introduced to obtain the characteristic data of mail behavior according to the pre-email header data processing.Compare the normal mail behavior with multiple discriminations.Through the verification of the system and the feedback result of the test data, it is proved that the design of the spam discriminant mechanism based on the user relationship behavior is reasonable and effective, and not only reduces the filtering load of the mail server, but also reduces the filtering load of the mail server.And has the high accuracy and the very good recall rate.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.098

【相似文献】