From the npr article:
Doesn't really match up with what you're saying. They have sophisticated software and don't seem to use much, if any, of the criteria you mentioned to determine if an account is real or likely not. In addition, Twitter is clearly aware there is a problem because they've challenged over a million different accounts for potentially being bots tweeting misinformation about Covid-19.
I don't know. The part about examining the number of followers and activity to determine if an account is a bot sounds exactly like what I was describing:
Researchers say they examine a Twitter account's followers, frequency of tweeting and how often the user is mentioned on the platform in determining whether the tweeter is a bot account.
As for "sophisticated software", they're still limited to the data that Twitter makes available from their API, which doesn't give them a tonne to work with. They wouldn't have IPs or sign-up emails, for example.
If the accounts are posting 50 tweets in 10 seconds, or a tweet every 10 minutes for 30 consecutive hours, sure, those are likely bots and that info would be available to researchers. If it is showing up as multiple countries in a short period of time, I'd want to rule out human users using a VPN, or at least know that they tried somehow.
That said, I'm sure there are bots on there. I'm just skeptical of this research because they get to first define what a bot is from the limited data they have and then report on how many they found.
They probably threw together some heuristic way of defining a bot. Probably something like high percentage of total tweets were about covid, with a negative sentiment score, maybe some references to keywords or low variability between tweets, maybe some time pattern that they feel is suspicious, etc. They don't have training data of "real" bot behaviour, which somewhat rules out traditional machine learning methods.
"When we see a whole bunch of tweets at the same time or back to back, it's like they're timed," Carley said. "We also look for use of the same exact hashtag, or messaging that appears to be copied and pasted from one bot to the next."
Nothing here screams 100% for sure bot behaviour. Seems like the way some people behave on twitter, especially when they're dufuses with an axe to grind about something political.
Basically, I don't think the research rules out the possibility that covid drove a lot of angry people onto twitter that normally don't tweet a lot. And those accounts would look a lot like bots. No followers, not following anybody, echo chamber behaviour, no conversations or friends on there, just pissed off tweets in large volumes all about covid. They don't look like regular human twitter users because they're not regular human twitter users. They're there to influence others about covid, not because they like using twitter. As this dies down many of them will disappear again from the platform, making the accounts look even more suspicious.
I'm not saying this theory is correct, but it seems as plausible to me as half the tweets representing espionage via bot army run by malicious state actors.
Maybe I'm wrong and they've got more exacting methods. It's hard to say without reading their paper, which unfortunately doesn't seem to be published anywhere yet.
