interesting spammer tricks

[David Relson]

On Wed, 24 Sep 2003 12:33:48 +0200
Przemyslaw Brojewski <przemek at my.tenbit.pl> wrote:

> On Tue, Sep 23, 2003 at 11:25:12PM -0400, David Relson wrote:
> > Tom,
> > 
> > I'm not sure exactly how good a foiler your message is.  The scoring
> > results using _my_ wordlists are given below.  Of interest is that
> > using my min_dev, robs, robx, etc the message parses with 11 hammish
> > tokens and 16 spammish tokens, which puts the message squarely in
> > the "Unsure" category (at 0.500000).  After using this message for
> > spam training (and then rescoring it), the score goes to Spam
> > 1.01e-5 (which means the score is 1.000000 - 1.01e-5, which is
> > 0.999995 or so).
> > 
> > Once again, a spam trick that works the first time.  After training
> > on the trick, bogofilter will nail it to the wall!!!
> > 
> > Time for a new, different spammer trick!
> > 
> It would be nice, if we could se how much scoring changed for
> other messages in training corpus. Just for completness.
> 
> Przemek.

An interesting idea...  I suspect most other messages aren't affected
since their tokens are different.  If you're feeling ambitious, why
don't you run an experiment and report back.

David