Bayesian filtering.

Spam, Spam, Spam - The first thing I do in the morning when I confront the onslaught of E-mail is to block highlight the entire list of new stuff, and then I unselect the things that look obviously like legitimate emails.  I suppose that every once in awhile I toss out a legitimate one.  But I only have so much time.  So if my E-mail in-box was a grazing field I have to tell you it's overgrazed.  If one or two cows die I can't help it.  I can't help it.  I don't have time to deal with spam. [from Ernie the Attorney]

Ernie seems mildly blase about his spam, compared to some attorneys I've dealt with, but his post appeared in my aggregator at about the same time as a few posts about Bayesian spam filtering (from Workbench was this post, which pointed to this article by Jeremy Bowers on iRights, which was followed up today with this today).   We have just started Bayesian filtering for a beta group, and it has gone well.

The few attorneys in our beta group have absolutely loved it.  The ability to train your own filter to recognize your own spam preferences has reduced spam for one attorney in the beta program from over 1000 messages a week to less than 10--with zero false positives thus far.  In other words, it's been great.  We certainly haven't rolled it out to everyone yet, but it has been significantly more effective at combatting spam that the filters we use on our firewall.  And most importantly, it has reduced the stress associated with wading through piles and piles of unwanted e-mail, and messages that our bayesian filter catches as spam are not forwarded to the attorney's handheld mail device (Blackberry, by Research in Motion).