[vox-tech] Is there better spam detection?

[Rick Moen]

Quoting Brian Lavender (brian at brie.com):

> It seems taht spammers have gotten keen to the Bag O Words Bayes
> analysis and now pump out some unrelated paragraph before they put in
> their spam.

Nah, that's a really _old_ spammer trick.  That's why you should never
leave a Bayesian classifier set to autolearn, and instead should feed it
spam and ham mboxes _you_ build with contents it'll benefit from.

> Is someone working on a meaningful spam analysis detection that I can
> use?

In my experience, your best results are from a multifactor approach,
with as much as possible of the heavy lifting being applied by MTA
rulesets during the incoming SMTP session.  You might want to use or at
least study J.P. Boggis's prepackaged setup for Exim4:
http://www.jcdigita.com/eximconfig/