Essentially: PageRank where nodes are words and edges are occurrence in the same document.

Really cool idea. Great work. I love to see this kind of stuff.