I use a plugin for squirrelmail that ads spam buttons. I also found a patch that modifies the behavior of that plugin to make it send the messages over to spamc to learn. In this way i managed to develop a very effective spam bayes database. I do not have much confidence in the auto-learn function because, in the past, it used to make big mistakes and it messed up the bayes database. The problem with training SA manually is to have a easy way to do that and i think this is better then letting SA decide what's spam and what not without letting the user train it.
My 2 cents over this matter :) > I'd like to ask this list before I jump into the regular SA list, but > has anyone else had the experience where the server learns way more > spam than ham? For example, yesterday my server learned 6533 spam > but only 111 ham. As a result, I'm getting a lot of ham with a > BAYES_99 score. I've been manually training SA with those messages, > but it's not a long-term solution. > > I recently upgraded from 3.0.x to 3.2.2, and I'm in the process of > upgrading to 3.2.3. > > Has anyone run into this before? Any good solutions? Right now, I'm > considering raising bayes_auto_learn_threshold_nonspam from 0.1 to > 1.0 (or even 1.5 or 2.0) so it starts learning more spam. I may need > to continue with my manual training for awhile though -- I came > across an old bug report and it appears that SA won't learn something > as ham if it triggered BAYES_99. > > -- > Tom Collins - [EMAIL PROTECTED] > Vpopmail - virtual domains for qmail: http://vpopmail.sf.net/ > QmailAdmin - web interface for Vpopmail: http://qmailadmin.sf.net/ > > >
