> Next steps after identifying the problem?

The only step I know that seems viable is the long, slow process of community education. Technical solutions don't really seem like they would help. I suppose we could put up a nag screen when there aren't many comments in a thread, asking people to check whether they're posting from the right place...but we've always tried to avoid that kind of thing, and even if each such step made sense individually they would soon compound into something annoying.

Fortunately, not everyone needs to understand these things, just enough to start feeding back into the system and affecting the culture.

What about a nag screen that only pops up when the post sounds snarky or angry, as determined by a ML/NLP model

I'd need to see such a model doing a good-enough job of identifying snarky / angry comments. The worst case of such a solution is that it's almost good enough but not quite—an uncanny valley type thing—in which case it will just piss people off. And anything less good than that would merely add noise.

Edit: of course, since comments are all public data, anyone who wanted to could work on such a model, and if anyone came up with one that was good enough, that would be significant work and I would very much like to know!

The expensive part is gathering good labels. Who wants to pay someone to tag things as snarky or angry?

We might be willing to pay. It might also be possible to marshal a community effort, if it were organized in the right way. Hmm - there might be something to this idea. The trick would be to set it up so that even if it failed, it would be an interesting failure.

You can get intriguing descriptive statistics (n-gram frequencies, time distributions) from even a relatively small labeling effort, well before you have much predictive value. So, in that sense you'll have something interesting in failure.

If you'd like a hand, please email me. I figure you have access to my hidden email address.