4 ms·
Something along the lines of: Comment scores for all comments added together (with negative scores taken off) / number of comments. A bit simplistic, and prone
by CM30 8y ago
Something along the lines of:
Comment scores for all comments added together (with negative scores taken off) / number of comments. A bit simplistic, and prone to echo chamber effects (high rated may just mean they say things they know everyone will agree with), but it's probably the most objective setup here, given the lack of miraculous super smart AI to 'judge' quality on any deeper level.
- tfehring 8y agoIn addition to echo chamber effects, this would strongly favor comments in popular threads. This may or may not be appropriate, depending on how you define "high-quality."
- zamalek 8y ago> this would strongly favor comments in popular threads Sum the total number of upvotes and score based on that. Track an independent score for downvotes. Display the two scores separately, in addition to the ratio between them. The MVP could be as simple as `votes_for_user_in_post/votes_in_post`. Scores would be intentionally low, you're looking for a track-record of greatness and not a once-off jackpot (or crackpot) opinion. There could be further improvements with population distributions: if most of the comments are receiving between 2 and 10 upvotes, a comment with 20 is truly exceptional (for that post). The problem that follows is that early comments that are mediocre will likely get more upvotes than late comments that are great. You're strangely aiming for the inverse of the HN comment ranking algorithm. StackOverflow has this problem: quick answers that are barely good enough typically receive more upvotes, and S/O hasn't solved that problem.