> > I would think that in the case of a company, you would be much better > > off with a global bayes database than individual bayes. > > > > I mean, unless everyone is getting all their own personal mail you have > > to think that the mails to the company will follow some sort of common > > pattern, no? > > Makes sense to me. So if I just leave it as on a company bases then I > guess > I setup a way for everyone to dump their emails into a folder somewhere > and > each day I manually review them to make sure they're getting them in there > correctly and that they are truly giving it spam etc. and then manually > run > sa-learn? Is that how other people are doing it?
IMAP is a good way of doing it. In my case, all the company mails classed as spam go to a single quarantine folder which I keep an eye on, and from there I weed out the ham emails and forward them on to their recipients. The spam emails get dragged to an imap folder on the mail relay which is manually sa-learned, and every so often I go through a selection of user inboxes and copy some emails into an IMAP folder to train SA on ham. Whether you do the latter may depend on the policy of your company, and applicable privacy laws. In any case, it works for me as I don't have to rely on any user intervention which, given the user base I deal with, is A Good Thing (TM). Ralph Pickering crownbc.com
