I explored whether this could be helpful in an online dispute resolution platform. The system could detect insulting or angry messages that threaten to derail a conversation, and suggest a more neutral way of formulating them. I think it's promising!
https://arxiv.org/abs/2307.16732