But discussion threads aren't articles. I don't believe they
can be summarized with fine-grained precision as if they were, because by definition the quality they provide is primarily to the participants, and terms like "interesting", or "meaningful" or "right" are entirely subjective.
You see this phenomenon whenever someone complains about why a particular story is popular. Any number of people have completely opposing ideas of what Hacker News should be. Everything one person finds interesting is completely noise to someone else.
You could apply user-level filters to the threads, I suppose - choose longer words over shorter, or longer posts over shorter posts, remove posts with words or phrases you object to. Hacker News could mitigate this single-channel effect by allowing tags or sub-boards, but I believe their solution so far is for the mods to manually tweak things when some stories get too popular. The site already puts high-karma posts at the top, which is supposed to act as a qualifier through consensus.
But to me, it seems the more you try to engineer the experience of discovery though automation, the less likely it is you'll discover things which might be interesting, but unexpected.