Ask HN: Open-source language model to identify bad online behavior?

I work for a startup that has a lot of UGC on it (primarily chat). We are ramping up our efforts to identify and act on hate speech, bullying, inappropriate sexual advances, and all the other bad things that people seem to do on the internet. Our tech team is starting to build a model of "bad" phrases and language, but I have to believe this work has been done somewhere and should be a public resource. Would appreciate suggestions!