Skip to Main Content
An interactive spam filter is proposed in this paper to reduce misclassification of spam. A set of weighted rules are applied to an email to decide if it's a spam. If a rule is triggered, the weight of it would be added. When the sum score of the total rules triggered is bigger than a threshold, the email would be classified as spam. The scores of rules are achieved by improved genetic algorithm. The improved genetic algorithm is applied to a corpus contains both legitimate spam and emails to get the initial score of each rule. During the filter process, end users' feedback is collected and used to adjust the scores of rules. Because retraining the whole data set is time consuming, the incremental learning algorithm is used to adjust rules' scores. The method is implemented and tested in one of the email servers in Network Research Center of Tsinghua University. The experimental result shows the false positive rate is reduced significantly by adding the user interaction to the spam filter.