By proxy, which is why there are false positives.

> How does one programmatically detect a flamewar? :)

An NLP/Machine learning model would be a great way! Probably also taking in other features as input such as the number of comments per second, and how deeply nested those comments are.

Let's build one!