Automatic Induction of Rules for e-mail Classification
by Elisabeth Crawford, Judy Kay, Eric McCreath
2001-12-07
Language:
English
Note: Proceedings of the Sixth Australasian Document Computing Symposium, Coffs Harbour, Australia, December 7, 2001.
External links
Full text: PDF
Abstract
Many users receive large amounts of email. Since a substantial part of that mail is kept for future reference, it is unsurprising that many mail tools allow users to create filtering rules that automatically do actions like saving mail in suitable folders. Unfortunately, most users do not make much use of this facility.
This paper describes the i-ems project which explores approaches to building a system to assist users in managing electronic mail. It does this by learning rules for classifying email so that it can assist the user in filing messages. This classification is also a precondition to automated action on behalf of the user and we are particularly interested in using it as part of a smart personal assistant.
We begin with a review of similar work and identify the core issues for the i-ems project. We relate the past work to classic information retrieval work in text classification and discuss the major new issues that we need to address. We present a simple user interface which automatically learns rules to improve the user's overall interaction with the email manager. We also present empirical results which compare four different learning approaches as email is progressively provided.
