On Tue, 10 Feb 2004, Kenneth Porter wrote: > This comes up regularly on the SpamAssassin lists. Another approach I'm > seeing lately is to use Bayes but assign high spam probabilities to tokens > that haven't been seen before.
That is not a good idea unless you have already built up a substantial Bayes database, and even then I wouldn't recommend it. Regards, David. _______________________________________________ Visit http://www.mimedefang.org and http://www.canit.ca MIMEDefang mailing list [EMAIL PROTECTED] http://lists.roaringpenguin.com/mailman/listinfo/mimedefang
