On Sat, Jun 06, 2015 at 12:48:42PM -0400, Michael B Allen wrote: > Right now I'm using procmail to put spam in user's .Spam Maildir > folder and then periodically just manually running sa-learn on > .LearnAsSpam and then deleting everything from both folders. This is > obviously not ideal since it's a) manual and b) deletes everything and > so there's no way to look into recent spam to look for false > positives. > > Can someone recommend a tool that handles this kind of post processing > in a sane way?
I use a single folder for fresh spam. When a user marks a message in this folder as seen, but still leaves it in the spam folder, or moves a message to this folder, it is considered "verified spam" and thus a candidate for learning. A second procmail script that is invoked periodically (quite often) feeds those "verified spam" messages to sa-learn (unless they have been autolearned as spam or have already been marked as BAYES_999), then moves them to a "spam-lo" folder (keeping the "read" mark intact) for reference and eventual deletion. I put very high scoring spam that is already known to the Bayesian filter (either autolearn=spam or BAYES_999) directly into the "spam-lo" folder. It's marked as "unread", so people can easily check if they miss anything or just want to make sure.
