zlacker

>>susam+(OP)
I will admit, one thing the crowd attention model does exceptionally well is surface the best comments on content. Whether it's HN, Instagram, YouTube, etc... the top comments are usually the "best", depending on how best is defined in the given context. On the silly Instagram meme videos my algo serves up, the top comments are invariably hilarious, often funnier than the actual content, and as you scroll it's impressive how the ordering by like count matches hilarity quite well.

>>gradus+kh
This works on platforms like HN, Less Wrong or niche subreddits, which

i) work on the reddit model (submissions + comments on them) ii) are heavily moderated (e.g. no memes but also specific restrictions like on a book series subreddit to not discuss the movie adaptations)

Then this vote-based ranking makes cream rise to the top, I agree.

In general, your "depending on how best is defined in the given context" does a lot of heavy lifting.