It would be interesting to build a model of specific HN users who downvote unpopular but true statements, and then use that to search for comments that may contain unpopular truths. Could be useful.
I guess we'd need access to data about how individual users vote, but it could be anonymized.