Judy Kay - Intelligent Electronic Mail Sorter projects
 Home    Bio   Research   Teaching
Intelligent Electronic Mail Sorter
Principals: Judy Kay, Irena Koprinska, Eric McCreath, Josiah Poon.
People currently receive large amounts of mail. There is considerable evidence that some of us are overwhelmed by this load, at times. For example, there are people who never seem to answer my email in a timely manner - not doubt, not you :-).
The goal of this work is to help people deal with the problem, not just as it stands, but as it will evolve. Part of the challenge is that people use email and associated mail management interfaces for a range of tasks: organising their workload, keeping reminders and records, managing projects ..... Although we might well be able to build elegant applications for some of these tasks, there is a real problem in getting people to use them. The current reality is that asyncronous communication via email is a multi-faceted and important communication mechanism.
There are many elements to the solution. Our current vision is to learn how to build systems that can help people organise email more effectively, perhaps to have some of it handled automatically. At the same time, we want the user to be in control of the processes and able to scrutinise the system to work out just what information the system holds and how it is used. We would also like to be able to exploit similarities between people to improve the process: then if one person acts on a piece of mail, this might be helpful in predicting how it would be handled by another person who has similar concerns.
iems 1
Elizabeth Crawford, Judy Kay, Eric McCreath
This project has provided foundations on two fronts. First, it has explored a novel interface, reported in the IUI paper below. This organises the inbox into predicted categories. The second part of this work has been the exploration of different machine learning algorithms for the task of predicting the way that the user will classify their email.
Acknowledgements: This work was funded by a University of Sydney Sesquicentenary Grant, 2001, McCreath and Kay. It has also received support from the SITCRC, Smart Internet Technology CRC in 2002.

Crawford, E, J Kay and E McCreath, (2002) IEMS - The Intelligent Email Sorter, Proceedings of the ICML International Conference on Machine Learning, 2002, 83-90.


Singh, S, J Kay, A Ryan, E McCreath, B Kummerfeld, (2001) Managing corporate email: connecting the social, policy and technical perspectives, Communications Research Forum, http://www.dcita.gov.au/crf.


Crawford, E, J Kay and E McCreath, (2002) An Intelligent Interface for Sorting Electronic Mail, Proceedings of the 2002 Conference on Intelligent User Interfaces, ACM, 182-183.


Crawford, E, J Kay and E McCreath, (2001) Automatic Induction of Rules for e-mail Classification, Proceedings of ADCS'2001, Australian Document Computing Symposium, 13-20. online proceedings


Kay, J and E McCreath, (2001) Automatic Induction of Rules for e-mail Classification, Proceedings of the Workshop on Machine Learning for User Modelling, UM2001, Sonthofen, Germany, 59-66. online version

Email as conversations
Jyot Boparai, Judy Kay
Some email constitutes a conversation. This project explored ways to enhance iems to support this. It made use of standard mail headers. In addition, when these were inadequate, it applied machine learning to assist. It also explored the modelling of senders so that the machine learning could be applied effectively for those users who tended to have behaviours that thwarted the simple mail threading based on standard mail headers.

Boparai, J and J Kay, (2002) Workflow Based Just-in-time Training, Kay, J and J Thom, (eds) Proceedings ADCS2002, Australian Document Computing Symposium, 16 December 2002, to appear. online proceedings

Current projects available:
At this stage, we are getting a good sense of how well machine learning is able to predict the way that a user will classify their email. We have not yet incorporated background knowledge about the learner into the learners. This is one important direction for future work.
There has been considerable work on distinguishing junk mail. We have a good sense of the level of accuracy one can reasonably expect for this two-class task (junk mail v everything else). However, we have yet to do a more thorough exploration of the relative power of different machine learning approaches for different categories. The aim of this project is to be able to inform the user how confident the system is about a predicted classification.
Multi-class email management. The existing work on iems has restricted mail to belong to a single category. In practice, people often think of email as fitting several categories. This can be either a project that concentrates on the interface issues or the machine learning.
Sender-based approaches to classifying and managing email seem promising. This project aims to build a user model for each sender address. A very simple version of this would involve the classic white-lists that many people use to manage junk mail. This project goes further, by keeping additional information about each sender on the white list so that this can be used to help predict the way that their mail should be handled. This involves making some inferences based on elements of the email address of the sender.
Top of Page