I do not agree with the arguments that a Twitter provided sample of the data would be enough. You would have to be sure that the sample is statistically repres

I do not agree with the arguments that a Twitter provided sample of the data would be enough. You would have to be sure that the sample is statistically representative of the overall population, which would require prior analysis with access to the whole data. Example of biases that you would need to adjust the sample for: - Maybe the part of active users receiving ads change shape with time, or is biased toward a certain type of user. - As more users are active on the site, a random sampling would overrepresent recent messages over previous ones. Or maybe the reverse if Twitter lost users. - Maybe %users being bots is not that interesting, versus %messages posted by bots. 5% of users being bots can represent 30% of messages being sent by bots. - If you're willing to train some ML model to detect bots, you're most likely going to need as much out of sample as you could have. - Etc