Igor Chudov wrote:
Bayes, applied to pdf spams, always classifies the message as ham and increases the score. It is not reliable for PDF messages where thecontent is in PDF.
Sounds like you need to train Bayes on those messages. Over here, Bayes is misclassifying less than 15% of PDF spams, and only a handful of those are getting through.
-- Kelson Vibber SpeedGate Communications <www.speed.net>