If the document were html, something like epubcfi might work better than regular expressions. This is what the epub people are recommending for a similar use case - sharing highlights and notes in epub books whose content may be updated.

For locating links/comments from people you don't know, I'd imagine a service like Google. Something that indexes the entire web and answers queries about which links overlay this document. I'd think they'd have to be ranked somehow. (Technically, Google already does this with "link:http://..." queries, but standard HTML links don't reference document fragments.)