Frequently you need to pick a single word, or perhaps two, from the title in order to find related articles. Using something like TF-IDF on submission titles (perhaps after mods have "normalized" them) then works well on the search. Then you want to make sure the articles really are the same, so you need to download the article, extract the relevant text, and do a similar sort of comparison.

Or just eyeball it. Or allow for mistakes.

Then there's the question of deciding whether it's worth having a cross-reference at all. Heuristics about number of comments, age of submission, etc, might be relevant.

In general, it's complicated.