Training a model on personal social media posts (text) might lead to privacy leaks, so I'd suggest against this practice (hopefully, it's not a thing, or too much of a thing).