From: Karl Anderson (kra@monkey.org)
Date: Mon Apr 17 2000 - 12:55:16 PDT
Eugene Leitl <eugene.leitl@lrz.uni-muenchen.de> writes:
> I'm looking for a _working_ intelligent, learning mail filter, ideally
> using neuronal nets.
Neural nets on what input?
> I want to create a special account, and sign it up to a large number
> of mailing lists/usenet newsfroups, yet still keep the volume
> manageable, and S/N tolerable.
I get fair results using gnus for usenet & mail. It's a user agent, I
could see it as a good user end of the chain.
For example, I wait until rec.bicycles.soc collects 400 articles, read
five to ten of them, and dump the rest. Pretty simple - seed it by
scoring regexps in the subject line (can search the body, but that's
slow), then it scores authors, subjects (or whatever you want to hack)
by your actions - read, kill thread, delete, catchup. So I end up not
reading people who've tended to get sucked into arguments, and I never
read a thread that I don't like twice, unless there's a good author in
there, & authors who contribute to threads that I kill get scored down
unless they also contribute to threads that I read.
Needs shared input a la Grouplens, tho, but that needs to be
user-agent neutral, & somewhat medium neutral.
-- Karl Anderson kra@monkey.org http://www.pobox.com/~kra/
This archive was generated by hypermail 2b29 : Mon Apr 17 2000 - 12:30:03 PDT