[Mimedefang] LZW, Gifs, and fingerprinting stock spams

[David F. Skoll]

Kevin A. McGrail wrote:

> I will also say that the stock image spams have been very effective at
> thwarting traditional anti-spam techniques.

Not to blow our own horn too much, but...

We've had pretty good luck with our RPTN system.  It's a shared Bayes
database.  A couple of hundred of our customer sites submit "votes" with
word-counts to add to a large shared Bayes database, which we update
and redistribute every night.  The non-image junk in most of those
spams usually scores very high in our Bayes implementation.
Our current RPTN database contains words and word-pairs from 451,506
spams and 224,318 hams, for a total of just under 7 million tokens.

Regards,

David.