I think Paul overrates the value of the global view on such data. Simple link count works well for evaluation of the importance of the web page until the Heisenberg effect kicks in (people start to adapt their pages to score better in the algorithm). My hunch is that it is enough to know the local structure of a densely-connected social graph.

Actually, link counts are pretty much worthless.

Really? Can you say why?

I haven't tried it, but link counts look like a reasonable first approximation of pagerank. I thought this is also the consensus among researchers, e.g.: http://anand.typepad.com/datawocky/2008/03/more-data-usual.h... http://research.microsoft.com/users/nickcr/pubs/craswell_tre...

It's easy for a low quality page to have a bunch of links. The blog post you linked is also wrong about AdWords -- the pricing model is important, but there's a lot more to it than that (which is part of the reason why Yahoo can't catch up).

I think another HN reader as well as myself mentioned this earlier: Amazon has a much richer dataset than Facebook, and the context of purchasing when you visit the site. Their recommendation algorithms have cost me so much of my money, I don't even want to think about it. The implied and clustered behavioral data for advertising is all there, whether people are "friends" or not. The only thing I see that they lack is a buzzphrase like "social graph".

Perhaps "social graph" is a natural phrase only if you use the word "graph" naturally. It certainly feels natural to me.