How is this computed? It looks cool, but it's difficult to use as a book discovery tool unless you know more about why it considers one book related to another.