I suspect that the way they came to that conclusion was: any post mentioning one of those groups, that also had a negative sentiment rating, meant that sentiment was directed at that group. Which is horribly dishonest. What’s more likely is someone to be angry (which registers as negative sentiment) about those groups being mistreated or what have you. By the naive approach they seem to have taken, that’s indistinguishable from being mad at that group.

Also, the methodology they describe, and the conclusions they come to don’t align. They don’t describe any methodology by which they could determine that the identities are being attacked. It would be like if they concluded some cause-and-effect relationship but their methodology had absolutely no way of establishing a causal relationship in the data.