That's really neat, had taken a couple stabs at this myself but not gotten all that far.

I think there's probably some really interesting data in the comments (maybe just take top ten comments, 3 layers deep?) And wrt following links, one idea that occurred to me was taking a screenshot of the page linked to and basing part of your model on that... I suspect there may be graphical, layout similarities in some of the pages people like.

But kudos for doing it, very cool!