SPAM filter comparisons
SPAM is a truly egregious problem. This article does some analysis of popular spam filtering software. It is no surprise to me that signature-based programs tend to have slightly higher false positive rates and that Bayesian filters work best when trained properly, and have low or zero false positive rates.
I use bogofilter and have a corpus of thousands of spam messages that I have received, going back 6 or more years. It works great. I don’t have to keep checking my spam bucket for false positives all the time and occasionally have to retrain if I get a false negative but online life is much better than before bogofilter. I was constantly having to scan through the spam bucket to pick out false positives.