[91720] in North American Network Operators' Group
Re: SORBS Contact
daemon@ATHENA.MIT.EDU (Joel Jaeggli)
Thu Aug 10 12:46:37 2006
Date: Thu, 10 Aug 2006 09:45:15 -0700
From: Joel Jaeggli <joelja@uoregon.edu>
To: "Derek J. Balling" <deballing@vassar.edu>
Cc: apoindex@aoc.nrao.edu, nanog@merit.edu
In-Reply-To: <980B8D1D-373F-4A09-9D0B-6B6EF04FD4B7@vassar.edu>
Errors-To: owner-nanog@merit.edu
On Wed, 9 Aug 2006 23:51:58 -0400
"Derek J. Balling" <deballing@vassar.edu> wrote:
> On Aug 9, 2006, at 10:59 PM, Allan Poindexter wrote:
> > At LISA a couple of years ago a Microsoftie got up at the SPAM
> > symposium and told of an experiment they did where they asked their
> > hotmail users to identify their mail messages as spam or not.
<snip>
> The recipient is
> > the only person who can determine these things.
Sure, but humans aren't perfectly accurate...
Early tests with bayesian classifiers, on the false postive rate, tended to indicate that building a classifier with a lower false postive rate than the humans was pretty easy.
Certainly my own experience is that I occassionaly tag things as junk, or mis-moderate messages to mailing lists. my own false postive rate is probably less than 1% spammassassain's is much lower than that. false negatives however are a reason I sitll have to tag things.
> I'm gonna hold up the "I call bullshit" card here. Recipients most
> certainly *can* get it wrong.
>
>
>
