Fingerprinting a Known #Misinformation Campaign: Breaking Down the Tactics of the Russian Internet Agency (IRA) in 2016. #SNA #networkscience #elxn43 #cdnpoli #Transparency

This post is part of the Social Media Lab #ELXN43 Transparency Initiative. It’s part one of a two-part post that summarizes our ongoing research on the role of social bots and trolls in our public discourse. The post is an abridge version of Dr. Gruzd’s recent talk, “FakeNews” Travels Fast — How Social Bots and Trolls Are Reshaping Public Debates, at the Dean’s Speaker Series at Ted Rogers School of Management on Oct 7, 2019. You can find the SlideShare here and the video here.

Today’s story begins in 2008. This is when a relatively unknown 1st term US Senator from the State of Illinois, Barack Obama, won his first US presidency. The success of Obama’s team can largely be attributed to their use of social media, which allowed them to share information quickly, reach many new potential voters and turn them into supporters and voters. Following Obama’s lead, in the early twenty-tens, politicians in Canada also started to actively use social media – so much so that the Globe & Mail declared the 2011 Federal election as Canada’s first “social media election.”

However, in just a few years later, in 2016, in his farewell speech, President Obama expressed concerns that social media was becoming a “threat to democracy.” This statement was ironic coming from a President who will be remembered, among many other firsts) as the “first social media President” with his own social media archive. But, his sentiment is understandable, as there is a lot of ambivalence about social media today. When we see its role in popular political uprisings or charitable fundraising campaigns, we recognize it as a force for good. And when we see it is used to recruit terrorists or propagate racism, we condemn it as a malicious, harmful tool.

The power of social media to serve as a medium for good or bad was on full display in 2016 during the last U.S. Presidential election and the U.K.’s Brexit referendum. This is when social media was weaponized to spread misinformation in an attempt to undermine democratic processes and influence people’s votes in those countries.

Social Network Analysis (SNA)

At the Social Media Lab, we are interested in studying how misinformation propagates through different online networks. We use a method called Social Network Analysis (SNA) to visualize and track how misinformation spreads on social media platforms such as Twitter.

To analyze Twitter data using SNA, we first need to represent it as a network. The first step in this process is to represent Twitter users (individuals or organizations) as nodes; and interactions among users (reply, retweet or mention) as ties between nodes. We can also assign a weight to each tie to represent its strength based on the number of exchanges between a pair of accounts. For example, a single tweet posted by Donald Trump to the @JustinTrudeau and @WhiteHouse accounts can be visualized as a simple network with three nodes and ties going from Trump’s account to the two accounts because both are mentioned in the tweet.

By applying the same process over and over again to a thousand (or even a million) tweets, we can visualize many simultaneous conversations as a communications network. This, in turn, will help us discover the most and least influential users in the conversations, to look for suspicious interactions and accounts, check if users tend to cluster around shared interests such as political views and determine why.

2016 U.S. Presidential Election

Next, to further explain SNA and make it more concrete, let’s apply the SNA method to a known misinformation campaign waged by the Russian Internet Agency (IRA) on Twitter during the 2016 U.S. Presidential election. Information about this political influence campaign was released by Twitter in 2018 as part of a massive dataset containing 9 million tweets and retweets from nearly 4,000 accounts associated with IRA over an 8-year period.

For the purpose of this post, we will focus on data from the 30-day period leading up to the 2016 U.S. Presidential election. By visualizing this sample dataset as a network, you will see how these accounts tried to sway political attitudes and who the Russians were targeting.

The figure below shows how political discussions on Twitter involving approximately 38,000 accounts appear as a network. While only 1% of these accounts were controlled by the IRA (represented here as red nodes), they generated over 200,000 tweets and retweets during this one month period. An example of an account associated with IRA is @TEN_GOP. It is currently blocked by Twitter, but when it was live, it misrepresented itself as the “Unofficial Twitter account of Tennessee Republicans”, gaining more than 145,000 followers.

It’s hard to know for sure how many of these accounts were ‘trolls’ (controlled by humans in Russia), and how many were setup as ‘bots’ to engage others automatically. More than likely it was a little bit of both. The fact that some of these red nodes are clustered together is a sign of coordination and that different clusters tried to engage different audiences on Twitter. To better understand the political interference strategy employed by these IRA-associated accounts, let’s examine each of the four largest clusters.

IRA Cluster 1

Accounts in the largest cluster were sharing pro- Donald Trump and anti-Hillary Clinton messages. This is something that you would expect to see in an election interference campaign operating in favor of a particular candidate.

IRA Cluster 2

But the network also reveals a second cluster of accounts that actively shared content using the #BlackLivesMatter hashtag.

This reveals the complex nature of the influence strategy being used. There was clear evidence that a large group of bots and trolls were used to support a particular candidate and/or to discredit their opponent. However, the presence of IRA controlled accounts in the #BlackLivesMatter hashtag shows that they were also interested in exasperating social fault lines in US society by encouraging polarization and incivility in democratic society. The IRA was likely counting on the fact that if people are too busy fighting with each other, they might not be paying attention to what else is happening around them.

IRA Cluster 3

Another prominent cluster that emerged from the IRA dataset was a cluster of pro-Kremlin and pro-Putin accounts, many of which shared content in Russian.

While this content might have not been directed at the American public; it was likely part of a larger information campaign to ensure that the perspectives of Russian political elites and Russian news organizations are well represented and defended on Twitter. Many of the IRA accounts in this cluster helped to make Russian news sources more visible and dominant on Twitter, especially those related to foreign policy issues and particularly related to Ukraine and the U.S.

IRA Cluster 4

Unlike other clusters, a fourth group of accounts primarily shared sports and entertainment-related content.

At first glance this might appear strange. Why would the IRA be interested in talking to sport fans? This is a classic “bait & switch” type of tactic often employed by scammers. To gain followers (victims) these bots/trolls accounts would start off by sharing funny and engaging content that has nothing to do with politics at the beginning to get people’s attention; once these accounts have acquired enough followers, they then wait for the right moment to slip in more politically motivated posts into people’s timeline

What we also found interesting was that all these clusters were connected to the official Twitter account for YouTube. This suggests many links shared within different clusters would lead people to watch videos on YouTube. This finding emphasizes the connected nature of different social media sites and how easy it is for information and misinformation to flow from one platform to another, taking different shapes and forms to adopt to the affordances and norms of different environments. For instance, what may start as a simple text message on Twitter, may then be reproduced as a visually-appealing meme on Instagram, and then repackaged as an engaging video on YouTube, all in an effort to make the content more appealing and ubiquitous

Part II will show our preliminary results of the analysis of recent #CDNPoli tweets during the 2019 Federal Election in Canada.

By: Anatoliy Gruzd and edited by Donald Patterson and Philip Mai