On Wed, 26 Nov 2003 16:02:05 -0800, Robert Menschel <[EMAIL PROTECTED]> writes:
> Before you go through the pain of intensive Bayes learning with all that > transfering around (and work needed to keep full headers and such), you > may want to just let Bayes auto-learn work. > > Widen the auto-learn limits to, say, -0.5 for ham and +15 for spam. That > should be enough to catch the worst of your spam, and maybe some ham > (2.55 generated a fair amount of ham at -0.5 and below, 2.60 tends to not > generate negative scores other than Bayes, which is ignored for this > purpose). You'll probably want to manually feed some of your personal ham > so Bayes has enough ham and spam to work with. Also, another related idea. Pipe all of your outgoing non-list non-automated email through SA to learn as ham. If someone replies to a message, its probably ham. And, you'll learn the text of the origional message if its embedded into the reply. Bonus points if you temporarily archive all messages by message ID, then feed the origional message in if you detect an outgoing message with an In-Reply-To header citing a message in the archive. Thats a good source of a lot of free ham. THen just throw in your spam archive for the spam side, or autolearn anything over 10, or probably do both. Scott ------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. Does SourceForge.net help you be more productive? Does it help you create better code? SHARE THE LOVE, and help us help YOU! Click Here: http://sourceforge.net/donate/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
