On Wed, 18 Jul 2007, Jean-Paul Natola wrote: > I'm getting creamed with these spams but they are getting through > due to; > > -2.6 BAYES_00 BODY: Bayesian spam probability is 0 to 1% > [score: 0.0005] > > Why is this happening?
Well, most likely two causes: (1) your bayes is mistrained, or (2) the spams contain sufficient text that looks like your legitimate message traffic. I'll repost the questions I posted here a bit ago in response to a similar question: (0) Does the text in the low-scoring spam, apart from the commercial pitch, look like legitimate message traffic that you would expect at your site? (e.g. if you're in the medical profession, it contains a lot of text about medical topics.) (1) How often does this happen? (2) How big is your bayes database? (use "sa-learn --dump magic") (3) How are you training your bayes database? Autolearn? Or manual? (4) If manual, do you keep your corpus around after you've trained it? (5) If manual, do you let nontechnical users train it without review? Finally, have you installed any of the SARE rules from www.rulesemporium.com? While not "fixing" a low bayes score, they may counteract it when bayes gets fooled. -- John Hardin KA7OHZ http://www.impsec.org/~jhardin/ [EMAIL PROTECTED] FALaholic #11174 pgpk -a [EMAIL PROTECTED] key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79 ----------------------------------------------------------------------- Where We Want You To Go Today 07/05/07: Microsoft patents in-OS adware architecture incorporating spyware, profiling, competitor suppression and delivery confirmation (U.S. Patent #20070157227) ----------------------------------------------------------------------- 6 days until The 38th anniversary of Apollo 11 landing on the Moon