Active Learning to Classify Email

While the technique of active learning has been applied successfully in improving text classification, its use in email classification has still not been explored. This paper examines several of the stateof-the-art algorithms for active learning with support vector machines as they are applied to email folder classification. We also introduce several extensions to these methods specifically designed to improve the quality of active learning when used for email folders. We evaluated the relative accuracy of these algorithms using a large publicly available email corpus. Our results show that current methods for active learning used in text classification work poorly for email foldering, but by taking chronological information, such as receipt time, into account, we can improve upon them significantly.