Ben reminded me about SpamAssassin‘s sa-learn tool recently. It trains the Bayesian filter that is part of (but not all of) SpamAssassin. I usually get between 100 and 200 pieces of junk email a day, and so I cannot live without SpamAssassin, but I usually get a couple false negatives that Mail.app catches.
Normally I delete my spam after giving it a cursory glance, but I saved up about 500 emails to run SA against. What surprised me was that after I went through to process of learning my spam boxes and my ham (non-spam) inbox my false negatives actually got worse! I wonder if that’s because I have 2500 pieces of ham and only had 1/5th as much spam to compare against. I’ll keep my spam around for sa-learn and hopefully things will improve soon…