Re: 'intelligent' mail filtering
Alex Malinovich <demonbane@the-love-shack.net> writes:
> I've heard of filters that can "learn" what is and isn't spam by having
> you feed it anything you consider spam. What are people's experiences
> with them? Are the useful/reliable? Any problems with false
> negatives/positives?
I have had very good luck with spam-stat.el, which comes provided with
modern versions of Gnus (5.10.x). Fairly few false positives, fairly
few false negatives, and I can attach it to groups that aren't
actually news groups with a little hacking. The downside is, it's
specific to Gnus, which is, well, a very different way of life from
things like Evolution.
I've heard other people say that Bayesian filtering on its own isn't
"good enough", and I believe it's implemented alongside the other
tests in modern spamassassin. It has the upside that it can pretty
easily be trained to recognize things like automated virus bounces as
spam. When I used ifile, I ran into problems where it would recognize
pretty much any HTML mail as spam (even legitimate mail from LookOut!
users with misconfigured mailers), but spam-stat has done better on
that front. That still does seem to be the source of most of my false
positives, though.
--
David Maze dmaze@debian.org http://people.debian.org/~dmaze/
"Theoretical politics is interesting. Politicking should be illegal."
-- Abra Mitchell
Reply to: