a less conventional method

a less conventional method

I've dabbled with a few of the programs in the article, but settled on an easier to use (for me) solution.

I access email through IMAP and I use mutt to read it. I have done this for a long time and generally resist change. The Bayesian filters available didn't integrate with this configuration at all. I would have had to use fetchmail or something similar and configuring mutt to train seemed complex to me.

I settled on using Thunderbird soley as a Bayesian filter/trainer. I just run in minimized and it takes care of almost all of my spam. In mutt, I move any undetected spam messages to a special IMAP mailbox and occasionally train thunderbird on the contents of that mailbox.

The nicest part is that when Thunderbird marks messages as "Junk" and moves them to it's designated IMAP mailbox for spam, the messages still show up in mutt in my INBOX as deleted (until I hit x to sync the mailbox). This gives me a chance to spot any false positives (of which there are very few). Also Thunderbird deletes the spam it identifies after 2 weeks automatically saving me the cleanup task.