Agreed, but I don't see how starting yet another site will fix the problematic lack of discussion.

One would imagine posting MORE to the existing site would solve the problem more readily than starting a new one with even less users.

Academic Hacker News may be suffering because the term academic in the name scares a lot of people away.

Maybe it's just solving the wrong problem. AHN only discusses CS, but I'm sure the discussion would not devolve if you included other subjects (ie Physics, Biology). A potential problem is that though you might have read something interesting, it might not quite fit.

Perhaps because the articles are dense and require time to read (a prof. told me to expect to take at least three hours to read an academic paper properly as an expert in the field), it would help a lot if the original poster would help jump start the discussion by talking about the interesting features of the experiment or study. I'm not sure if this can be forced though.

AHN (completely) aside, I had another idea for a news site that addresses a pain point of mine. I'm not sure if it belongs in this thread, but it seems as good a place as any. Many, if not most, articles in the mainstream media are horribly devoid of content, information, and context. An aggregator that posts stories and has a historian annotate them and add links to documents (especially primary source!) describing previous relevant events (sorted by importance and date) might be really cool and useful. I am stumped as to how to actually make it work though as if you do it with personnel, your labor costs would be enormous. On the other hand, there may be a ready supply of liberal arts doctorates who missed out on professorships (I understand there is a major discrepancy between the supply and demand). The real trick of this would be that if you managed to realistically rate historical events according to how they impacted the world, then you could sort your front page by both newness and importance score, thus mining the gems that are misplaced in regular papers. However, this would be susceptible to gaming by writers - which is unavoidable in the long run.

Automation would be much better. However, it becomes a classification problem along waaaay too many dimensions, and assembling a supervised learning dataset would be a giant task that I'm not even sure how to do as these labels can be very uncertain.

The rewards would be great if some sort of system like this could be devised on a reasonably sized scale.

Maybe the catchier name will help? Also it wouldn't simply be an HN clone, it would have features specific to its use case.

I doubt it. Perhaps the interesting discussion and activity comes precisely from the fact that this community does not suffer from programming tunnel vision.

One-dimensional minds are rarely interesting to converse with, no matter how far they go in that one dimension.