Future Computing Environments

CyberNag

Version 1.0
Gregory Abowd
Chris Atkeson

Goals:

We are interested in an intelligent agent approach to e-mail filtering (which we would eventually like to extend to all kinds of messages). As a starting point, here is a listing of some of the features we would want in an e-mail assistant NOW, with the hope that over time we will be able to figure out what to do with all of these requests.

Clean-up:

I frequently have messages in my box that need to be around for a little while (maybe a day or two, say a meeting reminder) but then can be harmlessly deleted. But I never get around to deleting them, so my mailbox fills up. Sure, I could be a bit more careful about cleaning up my mailbox everyday, but that is exactly the kind of mundane, but slightly intelligent task I would like to delegate.

Cross-referencing:

I use MH and I can link a single message to bunch of folders in order to make it easier to find the message later on. I'd like that filing to be done automatically and somewhat reasonably. I have thought of setting up something like a WAIS server to index my mail. I should just have to recalculate the index one a week or so, but then I still have to file the darn message so it leaves my main inbox.

Prioritized inbox:

It sure would be nice if I could have messages marked as important and occasionally I am reminded about them as long as they are still "open". Once a message is considered "closed" it should be filed for me.

Even smarter inbox:

I know there is a way to set up MH to handle incoming messages a bit better than I do. For example, it should be able to put some messages automatically into a particular place for me to browse later. In my way of thinking, I am starting to treat a lot of my e-mail as a newsgroup. Part of the problem with newsgroups (apart from the noise-signal ratio) is that I have to explicitly go look at newsgroups to find out when interesting news is there. I'd rather it all come to me as e-mail (I may be alone in that feeling, but I basically only want one interface). In fact, if I could find a way to incorporate my e-mail with a WWW browswer, I'd do it in a second. The WWW browser (netscape in my case) is beginning to become a personal information manager (PIM) for me.

Paraphrasing

Imagine we had a mailer that instead of receiving mail and displaying it to you on request just paraphrased the mail for you. Instead of seeing a header like:
 980  03/31 Gwen Baker         SRC Meetings[[Hello everyone. I have scheduled the SRC meetings base

in xmh, for example, I would see something like:

 Gwen Baker sending an announcement about SRC Meeting times this quarter

Whether the mailer showed me only one of those or several doesn't matter. What matters is that it is trying to digest the information and present it to me in a summary form, like any good assistant would. I can then decide what to do next (read it, reply, schedule meeting, etc.) and the mailer can learn from my actions.

Auto-Threading

Here is another problem I would like an intelligent mailer to handle for me. When I am away from my mail for an extended period (which could only be an hour some days). I end up getting lots of mail that has to do with the same topic. I usually read the messages in the order they arrived, and that's not the best strategy, especially if someone sends a message later that overrides the earlier message. My intelligent mailer should be able to determine which message I should see first and whether I need to see the other message at all. This kind of functionality is more along the lines of the mailer showing me the single most important message first, but think about how this could be achieved with the keywords approach and come up with some solutions.

Keywords

We talked about what kind of intelligent mailer one could develop. The first idea was to build a mailer that just showed the user the N most important mail messages. An awful lot can be done using keywords, so an initial idea was to be able to use some set of user-extensible keywords to rank mail messages.

What is annoying about e-mail? I have lots of mail and it's difficult to file it away in such a way that it's easy to find again. So it was clear from my frustration that searching for mail is important but that I don't want to spend the time using a mechanism such as folders to facilitate future searches. Keywords to the rescue again. All folders are to me is an indexing mechanism to help future searches. Well, if the mailer already had a reasonable set of keywords, what's to prevent it from automatically "filing" a message as soon as it is received. That way, all I have to do is read it (do some other action based on the message) and the message can go away. If I need to recall it, I should be able to ask for it according to the keywords.

We all seemed to converge on this idea of using keywords as the behind the scenes magic. So what are the issues?

  1. How do you determine a good set of keywords?
  2. How do you allow for the system to build up more keywords?
  3. Once you have the keywords, what can you use them for besides indexing?
Here are some keywords:
meeting
schedule
seminar
{any times, or dates}
{the year/month/date message was received}
{the sender of the message}
Abowds
software architecture
formal methods
software engineering
jobs
Call for Participation
CFP
conference
workshop
journal
review

The system would be able to suggest keywords based on word frequency (if it shows up in more than five messages, maybe it is worth keeping around). Is there such a thing as a keyword decaying over time?


Future Computing Environments Georgia Institute of Technology