[Bug 55835] Learn filter rules "automatically"

[Johannes Ballé]

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
         
http://bugs.kde.org/show_bug.cgi?id=55835         




------- Additional Comments From os xn--ball-epa net  2008-03-06 10:04 -------
This reminds me of an idea I had a few months ago. The whole process of email filtering could be done using supervised learning (basically, the same way a bayes spam filter such as bogofilter works).

Meaning, the user would initially just assign a few mails to folders (or at some later point in time create a new folder and move some messages to it), and the system would learn similarities between the messages in the folders based on their headers. Then, when a new message arrives, the system would do a classification and move the new message to the best matching folder. Of course, the user should be given a way to correct classification errors once in a while.

This scheme would be extremely easy to understand even for novice users, since all they have to do is set "examples" for the system to learn from.

I've had a very good experience with the bogofilter (only once in a while it yields a false negative, and I've had no false positives at all for about two years), so I can't think of a reason why a similar scheme for regular email filtering shouldn't work. What do you think?