改进的一分类支持向量机的邮件过滤研究

Filtering e-mail based on improved One-class Support Vector Machine

服务器端存在多个用户,且人们对邮件内容的理解和认可程度不同,因此邮件过滤中涉及到不确定信息的处理。就邮件内容来看,邮件过滤通常涉及到隐私,不利于大量收集样本并评价打分。因此提出了一种基于改进的一分类支持向量机的邮件过滤方法。该方法优点在于:(1)用户只需为不确定性很强的待区分邮件给出隶属度;(2)只需收集和训练一类邮件样本,便可以建立邮件分类模型;(3)把隶属度首次引入到1-SVM中,并且由隶属度的值的大小来确定惩罚因子的值。通过仿真实验验证了该方法的有效性。

Because there are many users in server,and users have different understand or admitting degrees for the content of e-mails,uncertain information processing is dealt with in filtering e-mails.From the content of e-mails point of view,filtering e-mails always deals with privacy,this is disadvantage for largely collecting e-mails and evaluating them.Filtering e-mail based on improved one-class SVM is proposed,the advantages of the method are（1）users only give membership degrees for uncertain e-mails which will be dealt with;（2）classing e-mails model is constructed by a kind of e-mail samples;（3）membership degrees are discussed in one-class SVM,and membership degrees are also used to decide punish factors.Simulation shows that the method is effective.