On Fri, Nov 03, 2006 at 02:13:38PM -0600, Mar Matthias Darin wrote:
> Chris Lightfoot writes:
> >On Fri, Nov 03, 2006 at 09:03:26AM -0800, Marc Perkel wrote:
[...]
> >what is the true false positive rate? nb you have to
> >measure this, you can't just infer it from complaints.
>
> That would be determined by analysing the logs, collecting the IP
> addresses, then doing a systematic lookup of each ISP's TOS/AUP in relation
> to dynamic/dhcp addresses being permitted to send mail directly
> subsequently bypassing the provider's desinated MX server.
no! you need to ask the recipient of the mail whether they
wanted to receive it. That is the only way you can tell
whether it was spam or not -- users don't typically care
about idiotic conditions which ISPs try to apply to them
or to other people (and rightly so).
In particular you can never measure the true false
positive rate if you do not allow users to see enough
information about mail to them that you've rejected to
decide whether they actually wanted it or not. Of course
to get an accurate measurement you then need to sample.
--
``Any time you skip a commercial, you're actually stealing the programming.''
(Jamie Kellner, of television station Turner Broadcasting)
