Hello everyone, To better train our Bayes I thought about setting up two popboxes (spam@ and ham@) on a local domain, to keep them hidden from the outside world. My intention is to bounce there (by hand) false positives and false negatives.
I have modified exim's configuration to drop all non-local mail coming to those two addresses. My crontab will launch sa-learn on those maildirs every hour or so. This way I can also keep a customized corpus of examples. My final step to make it work, is to have SA _not_ check/rate the mail going to those addresses. I know that Bayes would skip SA's headers but the Subject line would still have ** SPAM ** on false positives, and I don't think I want to train Bayes to recognize that as ham. Since I am pretty new to the wonderful world of Spamassassin, before clicking all of my pieces into place, I thought about submitting this plan to you and see if any flag gets raised. Maybe it's the oldest idea in the book but after a very minor search I didn't find any evidence of its popularity, and that made me think that maybe it's a deeply flawed idea and I can't see that for some reasons. Thank you for your attention. Cordialità / Best regards / Gxis la ------------------------------------------------------------------------ Manuel Giorgini <[EMAIL PROTECTED]>, Programmatore INTERLOGICA e-business solutions - http://www.interlogica.net Via Fusinato, 27 - IT 30171 Mestre VE - Italia - Unione Europea Tel +39 041 099 30 00 (6 linee r.a.) - Fax +39 041 504 11 72 ------------------------------------------------------------------------
