Email Activity Assistant
leveraging the hidden
structure of email
[ what / who & where / storyboard / publications / download / press ]
The Email Activity Assistant is a Thunderbird extension that
organizes email by activities, tasks, processes or workflows (pick your
favorite term). Email-based activity
management promises to give users better tools for managing increasing volumes
of email, by organizing messages according to a user's activities or
tasks. For instance, a consumer
purchasing an item from an e-commerce vendor may receive a message confirming
the order, a warning of a delay, and then a shipment notification. An email client that understands this
structure can save users from having to manually trawl or search through long
lists of unrelated messages to manage their activities.
Current email activity management tools rely on simple heuristics
(such as message threads) or manual intervention to label incoming messages
according to which activity they belong.
This makes them inflexible and overly demanding of user effort to
effectively manage large volumes of email.
EAA gives users an activity-centric view of their messages. Activities are displayed with a nested folder
structure. In the screen-shot above, a
user’s e-commerce activities are sorted by vendor, and then each activity is
depicted as a folder containing the messages relevant to a particular
transaction. A key feature is that this
activity-centric folder view is virtual; EAA does not modify the user’s actual
folders.
Users can manually assign messages to activities, give activities
meaningful names, etc. But a key
innovation of EAA is that it incorporates several algorithms for automatically
labeling messages with activity-centric metadata. Specifically, the algorithms described in [Kushmerick & Lau, IUI-05] are used to process structured
activities such as e-commerce transactions, and the algorithms described in [Khoussainov & Kushmerick, CEAS-05] are used to manage informal activities
such as meeting schedule or collaborative document editing.
EAA represents activities with the process model formalism introduced in [Kushmerick & Lau, IUI-05].
An activity is a set of messages, and activities are grouped into
types. For instance, the messages
related to a single EBay auction would be represented as an activity of the
EBay type. An activity type is a
finite-state automaton, where messages correspond to transitions. An activity at a given point in time is thus
represented as a path through its process model, with a message assigned to
each edge. The activity-centric metadata
associated with each message includes the activity to which it belongs, that
activity’s type, and the transitions to which it corresponds. This generic approach can accommodate both
highly structured activities (eg, e-commerce transactions) and informal
activities between people (eg, meeting scheduling).
We have prepared a sequence
of screen shots
which shows EAA in action (0.5MB PDF).
EAA is being developed by Nicholas
Kushmerick, Rory
Parle, Luke O’Malley and Rinat Khoussainov, a research team in the School of Computer Science & Informatics at University
College Dublin. EAA arose from collaboration with the IBM
Dublin Software Laboratory’s Centre for Advanced Studies; we thank Tessa Lau,
Mark Dredze and Alexander Trousov for helpful discussions.
A one-pager that describes EAA for the
IUI-06 Demo Session is available here. Here are some longer
papers:
N. Kushmerick, T. Lau, M. Dredze, and R.
Khoussainov. Activity-centric email: A machine learning approach. In Proc. American Nat. Conf. Artificial
Intelligence, 2006. NECTAR paper.
N. Kushmerick & T. Lau. Automated email activity management: An unsupervised
learning approach.
In Proc. Int. Conf. Intelligent User Interfaces, 2005. (Honorable mentioned for Outstanding Paper award.)
R. Khoussainov & N. Kushmerick. Email task management: An iterative relational learning
approach. In
Proc. Conf. Email and Anti-Spam, 2005.
C. Danis, W. Kellogg, T. Lau, M. Dredze,
J. Stylos, and N. Kushmerick. Managers email: Beyond tasks and to-dos. In Proc. Conf. Human Factors in
Computing Systems, 2005. Short paper.
R. Khoussainov and N. Kushmerick. Relational learning for email task management. In Proc. Int. Joint Conf. Artificial
Intelligence, 2005. Poster.
M. Dredze, T. Lau & N. Kushmerick. Automatically classifying emails into activities. In Proc. Int. Conf. Intelligent User
Interfaces, 2006.
EAA will soon be available to the research community for
non-commercial use. Stay tuned!
In the meantime, we have released the Bridge, a generic EAA component that other researchers might find
useful. The Bridge simplifies the
process of enabling Javascript code running inside Thunderbird to talk to arbitrary
Java code. All communication happens via
a lightweight protocol over HTTP. Our
intent is that the Bridge will make it easier to write “intelligent” email applications
for Thunderbird in Java. If your Java
code needs deep access to all of Thunderbird’s data structures, then you
probably want JavaXPCOM.
But if just need to send messages to your Java backend for classification/annotation/processing,
then the Bridge is probably all you need.
To try it out, download the Bridge (0.6MB ZIP archive), and following the instructions in readme.txt.
EAA has attracted a bit of attention in IEEE Intelligent Systems, Silicon Republic (twice), MIT Technology Review, ACM TechNews, and Technology Research News.
Nicholas
Kushmerick, 02.05.2006