I've had good luck with HTML::Parser, but I think what you're asking for could end up being infinitely complicated.

You want to end up doing a high level compare, not a line by line or word by word compare. If you have to handle anyone's HTML, that could be impossible. If you're trying to version your own HTML, that will probably be easier -- you can focus on just a few tags.

I think I'd start by comparing the structure of the document for changes, and see where a paragraph has been added or a table row has been deleted. From there, I'd look at the text within the parts whose structure has not changed. Boy, that's a really interesting project. Let us know how it turns out.