Out of curiosity, what did you use to extract the text content of the page? Was it based around html, or did you actually compare the topical content?