On 03/17/2006 08:57 PM, Bowie Bailey wrote: > Warren Howard wrote: >> Same problem for me (using Spamassassin 3.1.0). I keep feeding >> sa-learn the "Re: PharamaCFcy news", always the same style of message >> but each time slightly different, the drugs and prices are on the >> left and rubbish like this dacgvishJybzjjt is on the right. The spam >> mail itself is in html and the source has lots of this > > Example of obnoxious spam with lots of <span>'s removed. > >> in it and no matter how much I feed to sa-learn I keep getting a Bayes >> score of 00 (BAYES_00=-2.599). > > A BAYES_00 hit is pretty serious. It indicates that Bayes is > confident that this email is NOT spam. It looks like your Bayes DB > may be messed up. Take a look at your logs and see if Bayes is > auto-learning your false-negatives. > > And, stupid question, but... When you sa-learn these messages, you > are learning them as SPAM, right? > > You may also want to double-check that your sa-learn routine is using > the same database that is used to check during delivery. Make sure > you are logged in as the same user who is receiving the email. >
Thanks heaps, definitely some things that I should be double checking, I suspect the messed up database, I fed the majority of the 200 SPAM mails to sa-learn, however it took a couple of days for me to realise that I needed to do this for Bayes to become active and by that time some 300 or so HAM messages had been auto-learned by sa-learn. I assumed (most likely incorrectly) that they must have been real HAM. I think it's back to the Bayes drawing board for me! Thanks again, Warren.
