I recently presented my work on the Russian Twitter disinformation campaign at PolMeth 2020. Kudos to the organizers of the event - they put together an excellent online conference in very little time. I encourage you to check out some of the other posters, which are available on the the conference website.
My poster is below, along with some more detailed commentary. You can pop the poster out in a separate tab here.
This project started thanks to a suggestion from Deen Freelon at UNC (via my advisor) to look at the other half of the Twitter Election Integrity Dataset, which is primarily in Russian. These tweets were part of a disinformation campaign conducted by the state-affiliated Internet Research Agency (IRA) in Russia. While the disinformation campaign targeted at the 2016 US presidential election is fairly well-known, the Russian language tweets are comparatively understudied.
Early on in this project, a puzzle emerged. The Russian-language campaign, consisting of a little shy of 4.9 million tweets and well over half the IRA users, peaked immediately after the Russian annexation of Crimea and the US/EU imposition of sanctions on Russia. Excellent work by Jennifer Pan, Margaret Roberts, Gary King, and others has established that the Chinese state uses social media as a tool for censorship via distraction. Simplifying quite a bit, the argument goes that direct censorship is costly and difficult, but by flooding social media with irrelevant information the state makes it harder for citizens to see undesirable information. Unless you’re an activist, posts about protests, corruption, etc. might not come across your feed very often, and if your feed is filled with distracting content you’re considerably less likely to see those posts.
Keen Russia observers will already see the puzzle - the Russian disinformation campaign peaked during what’s known as the “Russian Spring”, a dramatic revival of Russian nationalism and almost overnight meteoric rises in approval ratings for Putin, United Russia, and the Russian government as a whole. While dissenters certainly exist, the Russian Spring is remarkable as a moment of collective effervescence - citizens’ opinions on Ukraine, the Russian state, the EU, and the US rapidly converged. United Russia and Putin, somewhat unexpectedly, found themselves in the strongest domestic position they’d ever been in. Given a political context in which Russians largely supported the official narrative on events and overwhelmingly approved of the government’s actions, why would they possibly want to distract their citizenry? In other words, if everyone likes what you’re doing, it doesn’t make sense to try to move attention to irrelevant topics.
Authoritarian governments certainly can do, and often do, the wrong thing - and that’s what I initially expected to find here. Instead, I found active participation in the discourse surrounding Ukraine, the EU, the US and sanctions.
After a long and computationally expensive tuning process (4.9 million documents, even tweets, is fairly big data) I settled on a 15 topic model. That in itself is somewhat remarkable - that 4.9 million documents could be explained reasonably well by 15 topics suggests a high level of message coordination. Five particularly substantively interesting topics emerged:
- Pushing the news: news stories the IRA was attempting to promote, largely through quote tweets and replies. e.g. “Thoughtfully written. The Russian people are sincerely worried about the situation in Ukraine.” ~ 12% of total tweets.
- Putin and Patriotism: tweets hyping up Vladimir Putin, using national symbols important to the regime, and general nationalist dreck. e.g. “Don’t get it confused: the government is the government, but the state is each of us!“; ~ 8% of total tweets. If you check the graph in the middle panel of the poster for this topic, you’ll notice a large peak in prevelance around the 70th anniversary of the Soviet victory in WW2, an important event for the regime.
- “Personal” Opinions: (Emphasis on the air-qoutes around personal!) General personal commentary on current events. While the Pushing the News topic is commentary on current events as well, this tended to not be comments on articles or particularly retweet heavy. I think of it as the IRA users playing the role of a politically-engaged citizen. ~ 6% of total tweets. e.g. “What do you think would happen? Ukraine would just ride off into the sunset?”
- America: Predictably, largely discussion of American santions. Also, plenty of negative news stories on things happening in the US (we’ve been generating plenty of those). Surprisingly, not a lot about the American election - probably because the Russian disinfo campaign had tapered off in favor of the campaign targeted at the US election. ~ 9% of total tweets. e.g. “Why is Russia feeding the US?” (referring to Russian grain exports to the US)
- Russia vs. the EU: Unsurprisingly, largely discussion of EU sanctions. Heavily pushing the narrative that EU sanctions are worse for the EU than Russia. ~ 7% of total tweets. e.g. “Russia hasn’t even felt a thing yet, but the EU itself can’t withstand the effect of its own sanctions!”
These topics track pretty well with events on the ground, which I take as an indicator of good model fit. There’s a couple interesting trends:
- The Pushing the News topic, the most prevalent topic in the data, drops off in the period immediately following the annexation of Crimea in favor of the Putin and Patriotism topic. Favorable news was clearly less of an emphasis for the IRA for February-March 2014 - instead they attempted to activate/amplify feelings of patriotism.
- I see something similar with “Personal” Opinions following the imposition of sanctions, this time with the topic dropping off in favor of the News topic, the America topic, and the EU topic. For this period, it was less important for the IRA to offer “personal” commentary on events than it was to promote the idea that not only would Russia withstand sanctions, they would backfire on the US and the EU.
My initial conclusion from this analysis is that the Russian government wasn’t using the disinfo campaign to distract, or at least not in the way the Chinese state distracts. Rather, they were reacting to events on the ground and actively engaging with the Twitter discussion surrounding them. It’s important to note that these discussions weren’t particularly manufactured or controlled: there really was strong Russian support for the regime and its actions. In other words, the goal appears to have been to amplify the surge of support for the regime, rather than create it or control it (both of which are nearly impossible tasks in the decentralized world of social media).
The results raise an interesting substantive question that I’ve been grappling with for some time: do distraction and engagement actually have differential effects? Even if (like me) you spend far too much time on Twitter, you still have limited time. If the IRA engaging in certain topics boosted their visibility on Twitter, users were inevitably distracted from other content.
Future work on this project aims to untangle exactly that question by looking at the effect of these tweets, particularly on emotional engagement. Distraction should simply distract, whereas participation should distract and have some effect on attitudes/affect.
A separate project will also look at the use of images in the dataset - something I think it is very important to explore. Working with images is much more difficult than working with text, but modern social media is a very visual phenomenon. If we keep ignoring image data, we’re discarding tons of potentially useful information.
If you have any questions or feedback, feel free to contact me! I am always happy to talk about this subject.