I used to use POPFile as a trainable Bayesian classifier to mark spam and ham. Worked well. I used to use a system that decreased the amount of email I read b

I used to use POPFile as a trainable Bayesian classifier to mark spam and ham. Worked well. I used to use a system that decreased the amount of email I read by 100x. I recently built another system that decreased it to zero. The last two aren't organization friendly. The Bayesian classifier might be your best bet.