Proceedings of the Sixth Conference on Email and Anti-Spam (CEAS 2009),
Mountain View, CA, July 2009.
While misclassified spam imposes a burden on end-users, the cost of false positives is much higher. Therefore, significant effort has been spent attempting to conservatively optimize this binary classification decision. While modern email filtering is quite effective, the sheer volume of spam implies that even high precision and high recall filters yield non-zero misclassification, i.e.\ no classifier is ``perfect'' against adaptable adversaries. This paper makes explicit recognition of this balancing act and argues for: i) removing the burden of perfect classification from the classifier; ii) separating classification and filtering tasks; and iii) a human factors approach to filtering. We present initial work on SpamGUI, an operational and publicly available embodiment of these ideas.
[ Return to publications ]