Disappointed to see there was no manual comparison of the automated results to the actual statistics of the tweets. The article mentions several times that both automated and manual tweet reviewing flagged more tweets from black people as hate speech but there was no attempt to quantify exactly how much hate speech each group generated. Tweets by African-Americans were 1.5x as likely to get flagged as offensive but, hypothetically, if it happened to be the case that African-Americans were 1.5x as likely to engage in hate speech neither the researchers or article readers would know about it. Not that that's the case, but do some due diligence.