An Email Filtering Method for Reducing the Misclassification Cost

摘要

In order to reduce misclassification cost of an email,this paper describes a new approach to filter spam based on the cost-sensitive learning method which uses the postprocessing methods of classification results,adapting-promoting algorithm and the thought of threshold filtering mail.The concrete step is to improve the mail classifier and it will give the mail an initial score,and then the classifier filters mails according to level of the score.Firstly,the classifier judges a mail and gives it the preliminary score.Secondly,the classifier extracts the users' interest and quantifies it as another score.What's more,the former two scores are weighted and a mail's comprehensive score is gotten.Finally,the classifier filters the mail according to the settings threshold.We consider user's interest in two ways.For example,the user would like to express user's interest,and the classifier can obtain directly it Initial simulation demonstrates the potential of this approach can input users' interest with the explicit method,and reduce the loss of classification.