基于内容的反垃圾邮件过滤器研究

本文论述了基于内容的反垃圾邮件过滤器的构造方法。介绍了邮件表示中的几个技术问题:特征定义和选择,以及特征权值的估计;探讨了以朴素贝叶斯、支持向量机和最大熵模型为代表的机器学习方法如何构造反垃圾邮件过滤器,并对它们作了简要的评价;介绍了几个常用的衡量邮件过滤器性能的指标。

A Survey of Content-based Anti-spam Filtering

The paper reviews several approaches to construct a content-based anti-spam filter. Technical details about the definition of features, feature selection and weight estimation for the representation of email projected into a vector space are provided. The discussion of some popular machine learning methods such as Naive Bayesian classifier, Support Vector Machine and Maximum Entropy Model which are always used to construct a content based anti-spam filter is on the way and followed by a brief comment on these approaches. Finally, several frequently used ways to measure the performance of email filter system are also presented.