I think I saw someone suggest using a multi-armed bandit algorithm with upvotes as the objective. I'd like to see that!
An algorithm like that would balance the exploitation of showing current highly upvoted posts first with the exploration of showing posts with few votes first, so that every post gets its chance while still penalizing spam posts once they're conclusively known to be bad.
An algorithm like that would balance the exploitation of showing current highly upvoted posts first with the exploration of showing posts with few votes first, so that every post gets its chance while still penalizing spam posts once they're conclusively known to be bad.