[EM] Search Engines & Election Methods?

[wclark at xoom.org]

I've noticed something recently: more websites are explicitly listing the
URL of another page, rather than linking to the page itself.

In part, I believe they're doing this to avoid increasing the ranking of
the mentioned website on search engines such as Google, which use linking
structure to determine positioning in search results.

Until recently, if you did a search for the word "Jew" on Google, an
anti-semitic website came back as the top result.  This is because a large
number of other webpages mentioning the word "Jew" had links to that
website.  Many news articles that mentioned this fact (particularly those
on anti-defamation websites) would only mention the offending website by
URL (if at all) so as to avoid linking to it and ultimately contributing
to the problem.

On the flip side, there's a well-known phenomenon of "Google Bombs"
whereby a group of people all place links to a given website on their
webpages (or in their blogs) and also mention a certain word or phrase. 
Google's Pagerank algorithm ends up highly associating that website with
those words, and you end up with cute results like President Bush's
webpage coming up first when you search for "miserable failure."

I was wondering if anyone had examined Google's Pagerank algorithm (or the
ranking algorithms of any other search engine) as if it were an election
method.  Some of the problems search engines are currently facing (for
which I'd bet they'd be willing to fund substantial research) are similar
to strategy issues and fairness criteria satisfaction issues that come up
in discussion here all the time.

Roughly put, Google treats the web as if it were a set of elections (one
for each word or phrase) with each webpage casting approval votes for any
other webpage at all.

It also combines elements of Candidate Proxy (or Steve Eppley's Candidate
Withdrawal method, or the method Mike Ossipoff described in November
2000.)  The magnitude of webpage X's vote is determined in part by how
many other webpages are casting votes for webpage X (and also on how many
webpages are casting votes for *those* webpages, etc.)

It's a very interesting algorithm, in that it elegantly handles voting
loops (A votes for B, who votes for C, who votes for A) and even includes
a random "noise floor" element to increase performance of the system.  (A
search for "Google Pagerank" will point you to as many detailed
descriptions as you like.)

-Bill Clark



-- 
Protest the 2-Party Duopoly:
http://votenader.org/