Re: [vps-mail] Bayesian AND BlackLists

[Sikaspam <sikaspam@xxxxxxxxxxx>]

Jim Smith wrote:
> required_hits           5.0
Wow, 5 is really low! This may result in a lot of false positives.
> I'm not sure if Bayes is working or not; I didn't see a reduction in missed
> spam. Here are my follow-up questions:
Look at the spamassassin headers. You will see the scores.
> 1. What do I put in here to have the Bayes filtered items separated from the
> other Spamassassin stuff?
You could do that with procmail by reading the headers.
> 2. Is there something else that I need to install (such as Bayes?) on the
> server in order for Bayes to work?
No, but it doesn't work until you train it and then it requires a few 
hundred messages I think. See below.
> 3. Is there a site that I can read up on Bayes? I'm not clear how it works
> but I'm assuming it compares the email to a database of some sort. I don't
> know what the "auto learn" is but wouldn't mind 'teaching' it if that is
> worthwhile.

Why not start by looking at the documentation for spamassassin? It 
explains in detail how to train the Bayes filter. If you do a search on 
Bayesian filter or something like that you will find the original paper 
that explains it and even gives the math. You can also look up popfile 
which is another proxy that uses Bayes and probably has an easier to 
understand text about the philosophy.

hth
======================================================================
This is <vps-mail@xxxxxxxxxxxx>       <http://www.perlcode.org/lists/>
======================================================================