Re: A new use for Spamprobe - Interestingness vs Ho-hum Boring
Darxus-5+gNnNaL/BD/[email protected] Wed, 7 Jun 2006 19:10:55 -0400
| Newsgroups | gmane.mail.spam.spamprobe.general |
|---|---|
| Message-ID | <[email protected]> |
On 06/08, John Carter wrote: > Interesting to Me vs Ho Hum Boring! Sounds like fun, I look forward to hearing your results. The math for bayesian filtering works for any number of categories (spam and ham being only two). So instead of first classifying your email as ham or spam, and then classifying your ham as interesting or boring, you could, in one step, classify your email as spam, interesting ham, or boring ham. I don't know how much modification it would take, but spamprobe could be modified to support an un-fixed number of categories. Or you could use CRM114, which I believe already supports an arbitrary number of categories (but is rather more involved to configure). Or you could just do as you said and run another instance of spamprobe for each email. > What fixed system resources does spamprobe use could screw up this > Plan for Ham? How could I hack around that. The obvious one is > the database. Any lock files, etc? I believe the -d option will handle all related problems. I've considered doing this, but never tried. Also, I filter most mailing lists out to folders before I do spam filtering, since most mailing lists disallow posts from unsubscribed addresses which eliminates basically all spam. Lists that don't filter in that manner get procmailed out to files after spam filtering. The other reason for this is bayesian filtering's preference for balanced input. -- "I would believe only in a God that knows how to Dance." - Nietzsche http://www.ChaosReigns.com