【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。更多还原

【Abstract】 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.更多还原