[Mimedefang] HTML mail

[David F. Skoll]

On Wed, 10 Dec 2003, Kenneth Porter wrote:

> For legitimate mail, I agree. Unfortunately, I've been receiving a
> lot of spam recently where the spam is in the HTML part and the
> plaintext part has an innocuous poem (different with each piece). If
> I strip the HTML, then the spam indicators disappear before I have a
> chance to submit the whole thing to Bayes.

Yes, this would be a problem.  I solve it by extracting and saving a
"Bayes Signature" (basically, a list of all the tokens in the e-mail)
before doing any stripping and relaying.  I can then train our Bayes engine
on the signature if I need to.  (This is our own Bayes implementation,
not the SpamAssassin one.)

Regards,

David.