As we may perceive: inferring logical documents from hypertext