邮件过滤算法的比较  

Comparative study on email filtering algorithms

作　　者：李文斌[1] 陈嶷瑛[1] 刘椿年[2] 刘泰峰[1] 

机构地区：[1]石家庄经济学院信息工程学院,河北石家庄050031 [2]北京工业大学计算机学院,北京100022

出　　处：《计算机工程与设计》2008年第17期4433-4436,共4页Computer Engineering and Design

基　　金：河北省科技厅基金项目(07213507D、06213558);石家庄经济学院重点基金项目(Y200606)

摘　　要：探讨了邮件过滤器的体系,介绍了6种常用的过滤算法。在4个公用的数据集上,利用3个评价指标对这6种算法进行了全面的比较。实验的主要结论有:集成过滤器的方法有利于提高过滤效果;与其它两个数据集相比,PU1和Ling-spam数据集用于评价过滤器时会得出相对乐观的结果;NB和k-NN的表现不稳定,对数据集的敏感程度较高。对于邮件过滤研究者了解、改进已有算法,提出新的算法有一定的参考价值。The architecture ofemail filters is given, and six algorithms for blocking spam is introduced. On four public available datasets, we do a full comparative study on the six algorithms by using three evaluation criteria. The main conclusions drawn from the experimental results are firstly that combing multiple filters can gain better performance than a single filter. Secondly, comparing with other two datasets, on PU1 and Ling-spam, a filter shows more optimistical results. The performance of NB and k-NN is greatly decided by datasets. The paper is significative for a researcher who wants to study email filtering, improve or design an email algorithm.

关 键 词：邮件过滤 电子邮件 过滤体系 垃圾电子邮件 数据挖掘 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]