That Sunday Times Russian bot "study". Beyond using "is the language or location set to Russian" as a method for IDing covert Russian state actors, which has never seemed *terribly* convincing to me, there's something else that jumps out as a question. 1/5
The dataset is 20,000 election-related tweets over a four week period. But the article says there were "thousands" of accounts. So that's... 2 or 3 tweets a week? But (based on their previous work) a major signal the Swansea team use to ID "bots" is *high-volume* tweeting. 2/5
And that 20,000 tweets include retweets of others. So it seems reasonable to suspect (absent researchers sharing data) that the UK election was a *tiny* proportion of the tweets/retweets from these accounts. Like maybe 1-2% of their output? 3/5
There are LOTS of bot accounts out there. Most aren't the Kremlin, they're just spam. And a thing they do is randomly RT popular tweets to seem human. So without knowing what *else* these accounts were doing, saying they were trying to swing the election is a huge leap. 4/5
And yeah, if you've got a bot network randomly RTing anything that does numbers, and it's during the election period, it'll probably lean towards Labour because they dominated social media in 2017. Any sample of spambots could show the same pattern. 5/5
More on this: the lead researcher says the 20,000 tweets came from 13,000 accounts. So these "bots" were averaging less than one tweet per fortnight about the election.
