关键词: 垃圾邮件过滤, 版本空间, 主动学习, 投票熵, 委员会投票算法

Abstract: A method is put forward in the paper which can get informative samples from unlabeled-sample pool with stepped way. The method which is based on query-by-committee algorithm increases the sampling threshold dynamically and it is in order to solve the problem of spam filtering online. Through the new method, the number of samples which is used for labeling and training is further reduced and the accuracy of classifier can remain stable. By experiments on Spambase datasets, the effectiveness which can improve efficiency of machine learning is certificated.

Key words: spam filtering, version space, active learning, vote entropy, query-by-committee algorithm