Yeah, I've been doing a combination of the auto-learning and manual learning. I have a site-wide bayes db set up (like it said on the wiki) and I have an email address for ham and spam that people redirect messages to. I've told people that they have to use the Action -> Resend method in outlook to send messages to these two learning addresses, and as far as I know they have been (but I don't have any way to verify that).
Whenever I get one of these messages I make sure to do the Action -> Resend thing and send it on to the spam@ address so that bayes will learn from it. Just didn't know if there was something else wrong w/my bayes db or if it just needs more training.. Sounds like it just needs more training :) Thx k -----Original Message----- From: Rich Puhek [mailto:[EMAIL PROTECTED] Sent: Friday, February 20, 2004 12:22 To: Kevin Hanser Subject: Re: Bayes Question Kevin Hanser wrote: > Lately I've noticed that there seems to be more spam than usual > creeping > through our SpamAssassin installation. I looked @ the message headers, > and I've noticed that on most of the ones that are getting through, I'm > seeing this: > > BAYES_00 -4.90 > > What does that mean? Do I have an improperly trained Bayes Database > somehow? Is there a way that I can tell SA to reset this one Bayes rule? > > Thx! > > k from 23_bayes.cf: "describe BAYES_00 Bayesian spam probability is 0 to 1%" In other words, the Bayes system thinks that there's close to 0% chance that this message is spam (put another way, Bayes is telling you "I'm 99 to 100% sure that this is NOT spam"), so SA reduces the score of the message. You want to re-learn anything like this as spam. You don't want to change the rule (SA is doing the Right Thing, Bayes thinks it's not spam, you don't want it flagged as spam) you just want to train your DB. Have you done any bayes training, or just let SA autolearn until the DB kicked in? If you've just relied on autolearning, you'll probably have a lot of these until you re-learn enough improperly classified messages. see: http://wiki.spamassassin.org/w/BayesFaq for info on learning... I've had good results with periodically running a bayes learn on the spam I receive. Lately, I've gotten lazy, so I've only been learning on spam and ham that's close to my threshold level of 8. Then I'll occasionally search my spam folders for messages that do not have BAYES_99, and learn them as spam (after verifying that they *are* spam). --Rich
