I would like to work with a machine learning expert to train a model on the existing set of Steem posts and see how well it can "learn" to pick out high quality posts.
It would have to have no knowledge of the author of the post (so it doesn't learn just to pick popular authors) but just the content of the post and the final payout.
Then if that is used to evaluate new posts it would be interesting to see how different its evaluation of what the payout "should" be vs what it actually is. This would tell us to some degree how much who the author of a post is affects the payouts, and might allow us to find unknown authors that make high quality posts more easily.
I would bet that machine learning could do a lot with this topic.
An algorithm that actually works to make higher quality appear in trending would be great, but it's quite a task from where I'm sitting.