Johnson, S wrote: > The only reason I ask about if I should "learn" the messages is that > my users have a hard time putting good email into the good email > folder. Everyone is quick to put in spam messages though. My filter > is getting about 50 to 1 spam to ham right now. Everything I've > read/heard states that I should try to be close to 1 to 1 for optimum > spam hits. If I add this into the learn then I'll be shooting up the > filter to close to 100 to 1 (or more). Should I be worried about > that? >
About that only thing you can do to get more ham learned, without invading your user's illusion of email privacy, is turn on auto-learning and adjust bayes_auto_learn_threshold_spam to something reasonably high since you have a user driven spam feed, and maybe adjust bayes_auto_learn_threshold_nonspam to be a little more free with what it considers ham (shouldn't need to tho).
