The Grumpy Editor's guide to bayesian spam filters
Posted Feb 23, 2006 13:54 UTC (Thu) by glouis
Parent article: The Grumpy Editor's guide to bayesian spam filters
Disclaimer: I'm a bogofilter developer and the original author of the bogotune utility.
Bogofilter works fairly well out of the box, with minimal training, as you saw. With careful tuning, particularly of the spam and nonspam cutoff values, and a rather larger amount of training, results like 0.5% false negatives and <1 in 150,000 false positives are attainable. See the papers at http://www.bgl.nu/bogofilter/ for details. The work involved is nontrivial but rewarding.
to post comments)