Three Million Tweets: Russian Information Operations in America

• In Event: Russian Disinformation Activities

Abstract

America’s 2016 Presidential election brought Russian disinformation campaigns into the public discourse after the release of the March 2019 report by the Department of Justice appointed Special Counsel Robert S. Mueller. In his report, Mueller identified the Internet Research Agency (IRA) in St. Petersburg, Russia, as a significant perpetrator of active measures to influence the American people and progress the Russian Federation’s political agenda. This organization's activities, however, were already well-known before the Mueller Report. In September 2018, National Public Radio (NPR) published a story claiming that on every social issue, the IRA “sought to amplify controversy by playing up conflict” by posting content on the social media platform Twitter, equally on both extremes of the issues, with the sole exception of Guns and the National Rifle Association. NPR claimed that the IRA was overwhelmingly Pro-Gun, with 77% of their Gun-related tweets being Pro-Gun and 11% of the tweets being Anti-Gun. NPR’s assessment was based on nearly three million tweets from 2015-2017 that were archived and organized by Clemson University professors Darren Linvill and Patrick Warren.

This study’s goal was to replicate the results of the NPR study and: 1) Observe the weight of effort or number of tweets posted by the IRA compared to other social issues; 2) Observe a significant leaning of Pro-Gun tweets in the data; 3) Determine language and content sentiment of Pro-Gun and Anti-Gun tweets; and 4) Identify patterns and significant spikes in posting compared to other social issues or notable events over the period.

A mixed-method approach was conducted with the Linvill and Warren data by combining 2,973,371 tweets from 2,848 Twitter handles into a single data set. Then, utilizing the Text to Data Process, the tweets were parsed into a data frame, building a corpus before being tokenized into words and assembled into a document feature matrix with duplicates deleted. Using artificial intelligence, keywords were identified and categorized into subgroups that were either Pro or Anti-Gun to find all Gun-related tweets. To compare the Gun topic with other social issues during the period, two control groups were selected, Black Lives Matter (BLM) and Immigration, and keywords and sub-group keywords were identified for each. Using a bag-of-words approach, the keywords were searched, time and date were registered in Posixct format, and all were aggregated based on date.

The study utilized the algorithm sentimentR, which classifies words as positive or negative. A sample from each subgroup was manually assessed for content and language accuracy to ensure validity. Finally, a cross-topic analysis was conducted using graphical depictions over time to determine significant increases in the social topic or subgroup tweets to aid in identifying trends.

The initial results of the study disprove the 2018 NPR results that the IRA was overwhelmingly Pro-Gun and that in other social issues, such as Black Lives Matter and Immigration, the IRA was relatively balanced by supporting both extremes. It did prove that between the three social issues observed, Guns did garner the most weight of effort by the IRA, but there was no drastic leaning of Pro-Gun tweets, as claimed by NPR. In fact, sentiment overall towards Guns was, on average, found to be negative. BLM, on the other hand, was far more lopsided, with the IRA overwhelmingly posting Pro-BLM tweets. Guns, like Immigration, were much more balanced. Essentially, if the NPR study substituted BLM for guns, this initial study would have found it far more accurate.

Finally, Guns-related tweets peaked just before the 2016 Presidential election. This also corresponds to an overall spike in Russian troll activity right before WikiLeaks released Hillary Clinton’s campaign emails. In addition, May through August 2017 saw the most significant amount of activity in IRA-related social issue tweets. The largest spike in Anti-Gun tweets occurred in mid-to-late May 2017, with the largest spike in Pro-Gun tweets during mid-July 2017. This is correlated to the greatest spike in BLM-related posts and Pro-BLM subgroup posts, both happening in mid-July 2017. Anti-BLM posts, however, were negligible throughout the entire duration assessed. Also occurring during this time was the largest amount of Immigration-related and Anti-Immigration subgroup tweets, followed by a large increase in Pro-Immigration subgroup tweets at the beginning of August 2017. Other notable findings included the exact text duplicated across different Twitter handles and the same Twitter handle playing both sides of the social issue. Further analysis of this topic will include increasing the number of social issues analyzed and utilizing machine learning to assess individual tweet content and sentiment.

Authors

• Rudy L. Novak, US Air Force Academy

• Martin Wendiggensen, Johns Hopkins