Learning to Filter Electronic Mail - MAGI

Terry Payne & Pete Edwards

With the increase in the volume of data on the Internet, there is a growing need for tools which can aid a user in retrieving and filtering this data. One approach has been the development of interface agents to assist users with applications such as electronic mail, USENET news or WWW browsing. By using Machine Learning techniques, a user-profile can be induced by observing and analysing a user's behaviour. This profile can then be used to identify and filter messages/articles/documents of interest to the user. We have developed a number of such agents, based on the same common architecture:

MAGI aids a user in sorting incoming electronic mail messages. By interacting with a modified version of the xmail mail tool, a user can send/receive mail messages and organise their mail box. Features extracted from messages are used by a learning algorithm to induce a user profile; this is used to predict (and automate) user actions on new messages.

We have investigated the performance of two different learning techniques within this system: CN2 (Clark & Niblett, 1989), and IBPL. Issues such as developing trust in the agent's recommendations have been explored, as have alternative methods for extraction of features from documents.

Publications:

• Experience with Rule Induction & k-Nearest Neighbour Methods for Interface Agents that Learn (Abstract)
  IEEE Transactions on Knowledge & Data Engineering, 9 (2), 329-335, 1997.
  T R Payne, P Edwards & C L Green

• Interface Agents that Learn: An Investigation of Learning Issues in a Mail Agent Interface (Abstract)
  Applied Artificial Intelligence, 11 (1), 1-32, 1997.
  T R Payne & P Edwards

• Using Machine Learning to Enhance Software Tools for Internet Information Management
  A Franz & H Kitano (Eds), AAAI-96 Workshop on Internet-Based Information Systems, WS-96-06, AAAI Press, 48-55, 1996.
  C L Green & P Edwards

Funding:

Internal MSc Project; UK Engineering & Physical Sciences Research Council (EPSRC) Studentship.