On Sep 30, 2014, at 11:11 AM, John Hardin <jhar...@impsec.org> wrote:
> How are you training your Bayes database? > > How much have you trained it? It requires a certain minimum amount of both > spam *and* ham before it starts evaluating messages. I have a significantly trained DB and I get the same thing... lots of these "short" spams that trigger BAYES_50. The problem is that they are sufficiently short that Bayes just doesn't have a lot to pick up on... there are some spammy tokens but very few are repeated from email to email, and a bunch of "indeterminate" tokens. Most of these emails are caught by my local URI templates, although the spammers add new subdomains every day so it's hard to keep up. Most of them are caught by my filter, though, thankfully... I only have a dozen or so per day that slip through with new subdomains. Not sure if there's any way to improve the Bayes categorization for these emails, but I do have similar issues of BAYES_50 most of the time for these. --- Amir
