Well first off, I like to write essays too, and I really have been enjoying the fact people here are way more willing to engage in longer posts.

I think you're into something with how humans empathize (kind of interesting to me my first response when someone tells me about an conflict is to try to reconstruct the other person's perspective). I think there's definitely a lot to the way people think less critically the more emotional they get

But to round it all off, smaller communities help, but really it's a matter of self-reinforcing social structures and the ways that social network mechanisms interact with them.

Outrage is the strongest driver for participation - so posts that incite the most outrage will get far more votes and replies in either direction. The outraged position will be far more likely to vote, while people who don't feel as strongly are less likely to do so to the same extent. That skews the metrics most algorithms use to rank them, and so they get more visibility.

As this goes on, the group will shift - the outraged people only need to be a fraction of the group to seem like they're the majority, and people put off by it are more than likely going to leave what looks like a total echo chamber (especially if people get nasty or personal)

The outraged group also starts to feel like their position is actually the average of the group (e.g. the silent majority), and they might shift even further, becoming more extreme - as people's beliefs are relative to their perception of social norms.

This cycle repeats until it becomes so polarized a moderate opinion is seen as extreme, and might be attacked.

It's a difficult problem to solve - the only easy metrics are going to be votes, comments, and maybe if people stay or leave after viewing. There's more complex systems that might work - such as using ai to score additional metrics based on content, or (an idea bouncing around in the back of my head for a while) by profiling the users to try to boost consensus opinions to compete with "outrageous" ones. Obviously, this is way more computationally expensive and requires complex code that few will be in a position to understand (even if it were open source). These strategies could also be used to drive engagement or ad conversation at the expense of mental health (something that seems to be at least explored by some social media companies)

But small groups help in a very simple way -only so much media fits on a page. Even if the top comments are pure outrage porn, the other voices won't be buried

The other solution is moderation (it's in the name) - effective moderation of the tone and "rules of engagement" can tamp things down. But people generally don't like to be censored, and it doesn't scale - moderators are individuals, and too much to go through or dividing it up between larger groups of mods strips the nuance out of the process