一种有效的垃圾邮件过滤新方法

New effective method for spam filtering

受到信息粒度原理的启发,给出了一种有效的垃圾邮件过滤新方法。该方法训练过程是将训练样本集合中合法邮件类和垃圾邮件类拆分成四个小类,得到四个小类的类中心向量,从粒度原理角度来看,就是采用更细的粒度来描述训练样本的先验知识。过滤过程则将新进来的邮件分别与四个小类的类中心向量进行相似度比较,最终来判定所属类别。在公共垃圾邮件语料库上测试新方法,同时与目前过滤性能较高的KNN方法进行比较,结果显示新方法具有过滤精度高,过滤速度快等优点。

A new effective method for spam filtering according to the principle of granularity was presented. First, this method divided spam class and legit class in train corpus into four small classes, and four center vectors were obtained. In the view of the principle of granularity, smaller granularity is used to describe knowledge in train corpus. When faltering, the new E-mail was compared with four center vectors respectively to decide which class it belonged to. This method was tested on spain corpus and compared with KNN. The results show that the new method has some advantages including high accuracy, high speed of filtering and so on.