-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

John Andersen writes:
> > The note attached to that wiki entry that the method requires that
> > spammers not conciously change their writing style across text is
> > somewhat inaccurate. It takes a rather LARGE change in writing style,
> > and some identifying features will always be present. 
> 
> But this assumes the spammers are spamming for themselves.  Most
> of the ROKSO spammers are simply performing a service for their
> customers, and all they need do is let the customer write the text.
> 
> Mere knowledge that such analysis was being done (ya, they read
> this list) would trigger use of other text authors.
> 
> But assuming away any such objections, what could you do with this
> knowledge?  The analysis is not likely to be useful in court, so its
> usefulness would have to be merely as another SA test.

Well, that's not entirely true.  The analysis *may* be useful
in court.

> But do we have to go that far? Isn't that quite a stretch to
> find yet another identifier, when the current methods are working
> so well?
> 
> The single most successful new feature (IMHO) has been SURBL, 
> not only because of its effectiveness in identifying spam, but also
> because it takes away the economic incentive to spam, by 
> catching anything that even mentions URLs used by spammers.
> 
> We already know who the ROKSO people are, and that knowledge
> hasn't slowed them down one iota. 

I agree it probably would not make a good spam sign -- hence
I think it'd be mainly useful as a side-project.  And there
*are* spammers that evade even ROKSO.

Mind you, those papers are very keen on the idea that it *would*
be a good spam-sign. ;)

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFBEHmTQTcbUG5Y7woRAskuAKCMnwh4fJ6oaQRIEhfs3DvCRhfh9ACZAW4k
AJw5OVL6Je+PJmSRUKRqnRo=
=iEEn
-----END PGP SIGNATURE-----

Reply via email to