On Thu, Aug 12, 2004 at 09:54:14AM -0500, Ryan Malayter wrote:
|
| Has anyone ever seen an effective "reputation system for reputation
| systems"? I've seen a few statistical comparisons of DNSBLs, but they
| all seem to ignore the most significant item when it comes to spam
| filtering: the false-positive percentage. Of course, you have to have a
| person manually evaluating whether or not each message *is* spam to get
| this data, so perhaps that's why.
I believe the generally accepted guide to DNSBLs and their
ilk is at
http://www.sdsc.edu/~jeff/spam/Blacklists_Compared.html
| Such "reputation system grading" is probably something that is going to
| be even more necessary as SPF/MARID grow in popularity. People will need
| to know which reputation systems they can trust, right?
I am working with a number of industry leaders on developing
a rottentomatoes for domain-based reputation and
accreditation system. My initial thoughts are documented at
http:/spf.pobox.com/aspen.html.
Things have moved fast since I wrote that document.
Cloudmark has announced
http://rating.cloudmark.com/senderid/ (a domain-name-based
reputation service) and Verisign has announced its VDL (a
first approximation to domain-name-based accreditation).
In addition there is Gossip: http://www.sufficiently-advanced.net/
So it looks like it's going to happen.
The ultimate communities-based "consensual subjectivity"
spam scoring systems will probably end up with an
Amazon-like "recipients like you thought this message was
spam". The privacy implications are left as an exercise for
the reader :)