邮件过滤中一种改进的特征选择方法研究  

Improvement of Feature Selection Algorithm in Spam Filtering

作　　者：林伟[1] 柳荣其[1] 徐熙[1] 

机构地区：[1]西华大学数学与计算机学院,四川成都610039

出　　处：《计算机技术与发展》2009年第1期84-87,共4页Computer Technology and Development

基　　金：四川省青年软件创新工程基金(2007AA42);西华大学科技创新基金(07226161)

摘　　要：特征选择是基于内容的垃圾邮件过滤的重要过程,它有效地改善过滤效率和精度。文中分析了IG和CHI应用到垃圾邮件过滤中存在的不足,设计了一种改进的评价函数。而由于这种评价函数的特征选择方法只度量了特征与类的关系,忽略了特征之间依赖的关系,因此进一步用基于互信息的聚类方法来消除特征间冗余。实验结果表明该特征选择方法有效地提高了邮件过滤的性能。Feature selection which is a very important process for content- based spam filtering. It is very efficient to improve the efficiency and precision of filtration. In this paper analyzed and improved disadvantage of IG and CHI which is applied to spam filtering. But this evaluation function only evaluate the relationship between feature and class, they neglect the relationship among features. So used clustering to reduce redundancy which is measured by mutual information among features. The experimental results show that the improved approach is effective to spam filtering.

关 键 词：邮件过滤 特征选择 信息增益 CHI统计 聚类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]