Re: On PR and 'community building'

On Thu, Feb 02, 2006 at 05:16:34PM +0000, Robin Green wrote:
> On Thu, 2 Feb 2006 15:33:20 +0100
> [email protected] (Tomas Zerolo) wrote:
> >   - I wonder whether some of the techniques used for fighting spam
> > could be applied to this. Would it be possible to recognize the same
> >     writer (in different disguises) based on tri-gram (n-gram)
> > frequency distributions? What about a karma system à la slashdot
> > (but much more diverse: GroupLens comes to mind)?
> 
> Baaaaaaad. Spam is easily humanly-recognisable by at least 20% of the
> net population, so a spam-filter is humanly verifiable to ensure it
> doesn't block too much ham. This technique would be difficult to check
> up on because you wouldn't in general be able to be sure what was
> "spam" and what was "ham".

Worse: such systems are entirely reliant on being provided with
correct training data in order to function. And you can't really get
that either.

Also, spammers have no real motive to evade most anti-spam systems,
since their targets are the sort of people who don't know they need
one. In contrast, these PR goons are highly motivated to evade such
checks.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)

iD8DBQFD4mjmlpK98RSteX8RAs8vAKCDgjCowMENP/byQ73ujsBR8XxsfACeLtMG
PhTIaU7lIUuLOLUVL4Vtn6M=
=+a52
-----END PGP SIGNATURE-----