Incase any Ai / tech minded people are reading, this is a relevant quote:

> As fall semester kicked up, RoBhat hand-picked 100 Twitter accounts with automated behavior to serve as “ground truth” data to train their model. They picked accounts with several red flags: ones that joined the site, say, a month prior but had tweeted 10,000 times, or ones that were followed by thousands of other suspected bots.

While I was skeptical they could train a model without actually knowing for sure which accounts are bots, their assumptions here seem to make sense. There is a certain "tweets per day" ratio that can't be easily maintained by a human.