I came across an interesting article on using Bayesian networks to filter out spam messages. The probability that each message might be spam is calculated based on the frequency of words found in previous messages. It was an interesting read for me after taking a number of AI classes at the University of Michigan. After reading this article, somebody implemented a qmail filter in Perl using the algorithms outlined in the article. I wonder if the junk mail filters that hotmail and the Apple Mail tool use take advantage of Bayesian networks.I’ve also heard about a package called SpamAssassin that is supposedly very good at filtering out spam. Among other methods, it uses Vipul’s Razor, a package that compares hashes of the mail of all of its users to determine mass mailings.