>
Wouldn't a emotional keyword filter make for better statistical flamewar detection?Something like http://www.paulgraham.com/spam.html but fine tuned to flamewars? Perhaps. The mods have some easy automatic criteria, the flags, vouches and upvotes from users, and then apply manual moderation on top of that.
It may be hard to distinguish a good discussion about a war from a bad discussion about icecream. Or replace war and icecream with your favorite topics.
> Also, i guess length of comment might be a strong indicator of thread degeneration.
I like that. Very short comments are usually bad, but huge wall of text too. Perhaps they have something more advanced and never told us. It may be part of the secret sauce. (Or it may be good that they pretend they have advanced stuff, so people behave better.)
> Finally you could create fields of expertise for every account (upvoted comments keywords) and ban threads with non-experts majority commenting.
It's an interesting idea, but my guess is that it's too hard. I'm not sure which tags apply to me. In some topics I'm mathematician, but I comment a lot in physics stuff because I have an unfinished degree in physics. But it depend on the area. In some areas I know a lot and in others I can just skim the article and look at the graphics. [1]
My guess is that non-experts are always the majority. For me the important part is that the comments by experts float near the top. There are also some interesting comments from non-experts, sometimes with a good reply from an expert. Upvoting good comments is very important to make the discusion better.
[1] Protip: Reading medicine studies, remember to Ctr+F "exclusion", because sometimes after the study started they excluded some of the subjects, like the guy that had 100% success after excluding the 1 dead and 5 other weird cases.