Kremlin’s : Russian Tweets on the US and French elections.

Alla Baranovsky

April 1, 2018

Abstract The possibility of Russian meddling in the American elections in 2016 has made headlines around the world. As a result, the French election of 2017 was being watched with much trepidation for signs of a similar occurrence. But how do Rus- sian users of social media feel about the two elections? I address this question using a newly collected database of Russian tweets about the two elections. As the Rus- sian authorities have made no secret of their own increased reliance on social media platforms like Twitter for propaganda,1 I also gauge the presence of the pro- govern- mental narrative about both elections on Twitter. For instance, a strong predictor of favorable sentiment for Kremlin-approved candidates is the number of re-tweets for a status (indication that it is being mass distributed), as well as it originating with Today, a Kremlin supported news source. This indicates that Twitter, once an online space for the Western oriented, young opposition, is becoming appropriated by the regime’s propaganda machine, to accomplish both domestic and international goals.

Introduction

In March of 2017, a former FBI agent described his understanding of how Russian-created Twitter bots masquerading as Midwestern Republican voters spread fake news on the social media platform, promoted the name of and denigrated

1Ivan Sekretarev, “Kremlin’s ’Patriotic Education’ to Target Russian Kids Online," The Times, 30 April, 2017. Found at https://themoscowtimes.com/ news/kremlins-patriotic-education-targets-- schoolkids-online-57578. Last accessed 8 May, 2017.

1 his opponent, Hillary Clinton.2 Russian bots that pretended to be American voters, like the infamous “Jenna Abrams" account3, were not the extent of the Russian informational campaign during the US election. Russian bots and employees of the also propagated a large number of messages about the US election in Russian, on the Russian subsection of Twitter. The attempts of the Russian government to influence the outcome of the US election is an example of a sovereign state engaging in a propaganda campaign that is designed specifically to meddle in other countries’ domestic political affairs. Although in some cases there might be a direct, observable, benefit to the offending state’s actions, in many cases this benefit is unclear or obscured. That is, specifically to the Russian case, the attempt to elect a sympathetic president, who might be amenable to removing sanctions, has a clear and observable payoff, but the same country’s attempts to foment secessionist sentiment in the US is less easily explainable. Whatever the reason for engaging in informational attacks, the state that does it in the twenty-first century certainly has more channels and mechanisms for a successful campaign than before. Technological progress in general, and the rise of social media in particular, have changed the methods used by offending countries. Whereas in the USSR, the ideological precursor to Putin’s Russia, the main vehicle of propaganda was the news- paper, and later the TV, Putin’s public opinion army increasingly relies on media such as Twitter and . Kremlin’s “troll farms" (the Russian counterpart to the Chinese “50 cent army") reportedly pay hundreds of employees to create and promote hashtags,

2O’Connor, Gabe, “How Russian Twitter Bots Pumped Out Fake News During the 2016 Election," NPR, 3 April, 2017. Found at https://www.npr.org/sections/alltechconsidered/2017/04/03/522503844/how- russian-twitter-bots-pumped-out-fake-news-during-the-2016-election. Last accessed 27 March, 2018. 3O’Sullivan, Donie, “The Kremlin-linked troll Twitter can’t seem to shake," CNN, 29 November, 2017. Found at http://money.cnn.com/2017/11/28/media/jenna-abrams-russia-twitter-back- again/index.html?iid=EL. Last accessed 27 March, 2018.

2 among other things.4According to anecdotal evidence provided by journalists’ investi- gations, employees in such organizations spend their work days engaged in “trolling" – aggressive stalking and/or harrassing users of the social medium who are expressing an ideological opinion divergent from that of the Kremlin.5 Moreoever, some states, like Russia and China, already have sophisticated domestic propaganda resources that allow for an easy transfer of know-how to influencing public opinion abroad. Hybrid political regimes, like Russia and China, pay exceptionally close attention to domestic public opinion because stability of the incumbent, especially in moments of crisis, depends on elite calculations which, in turn, depend on public opinion. For instance, if a weakening authoritarian incumbent is caught in a corruption scandal, this might provoke large scale protests, as was the case in Russia in 2011 - 2013. Such a wave of open dissent causes elites to recalculate their chances of survival in power with the current incumbent, triggering a mechanism similar to a bank run. In some cases, this dynamic might result in the removal of the incumbent. This is why such regimes have for decades developed and maintained sophisticated mechanisms of manipulating the opinions of their citizens.6 In this study, I explore how Russian users of Twitter, previously composed almost exclusively of Western-oriented, urban, educated elites, but now also inclusive of Kremlin propagandists, talk about the US and the French elections of 2016 and 2017, respectively. This paper proceeds as follows. After reviewing existing theory on social media in politics in general, and in Russia in particular, I provide some qualitative evidence for Kremlin’s

4Andrei Zakharov, Polina Rusyaeva, “Rassledovanie RBK: kak iz ’fabriki trollei’ vyrasla ’fabrika me- dia,"RBK, 24 March, 2017. Found at http://www.rbc.ru/ magazine/2017/04/58d106b09a794710fa8934ac? from=subject 5A famous investigation that started further inquiry into Kremlin’s “troll farms" is Adrian Chen, “The Agency," 2 June, 2015. Last accessed 8 May, 2017. 6Henry Hale, “Eurasian Polities as Hybrid Regimes: The Case of Putin’s Russia," Journal of Eurasian Studies 1:1, 2010, pp. 33 - 41.

3 official position on presidential candidates. In the next section, I examine a database of tweets obtained via Twitter’s API, using select hashtags, in the days leading up to, during, and after the two elections. I look at frequencies of words to find the dominant narrative and the most talked about candidates. I find that during the US election, Donald Trump is talked about considerably more than Hillary Clinton. Similarly, Marine Le Pen is the most frequently mentioned candidate around the time of the French election. I also do a simple evaluation of sentiment by coding tweets as Kremlin-friendly (compatible with Kremlin-promoted narrative), Kremlin-unfriendly, or neutral/unclear. I find that the first category dominates during both of the elections. Finally, I examine the determinants of support for Kremlin-promoted candidates, and find that the number of retweets for the status, as well as its origins at Russia Today, a Kremlin-funded news source, are strong predictors. The final section concludes. In this paper, I do not consider another type of meddling in another nation’s political outcomes, which is arguably a more direct way to interfere in elections, that Russia has been accused of. Cyber meddling, hacking and leaking emails and other files, are out- side of the scope of this paper. Both the Clinton and the Macron campaigns, those of candidates antagonistic to the Kremlin-preferred scenario, have reported digital break-ins with evidence pointing to Russian hackers. Although I do not address this aspect of elec- tion meddling further, it informs the general context of Russia’s attempts at influencing a political outcome abroad. Why is it important to examine the tweets, since they are not in- tended for the actual population of the US voters? This is because examining Russia- originated tweets in English requires knowledge of which Twitter accounts are registered to Russian users. In the absence of such information, the Russian Twittersphere approx- imates the sentiment that the country’s propaganda machine would want to disseminate

4 abroad. The presumed intended mechanism of the process that I am focusing on in this paper, the dissemination of a Kremlin-promoted narrative on Twitter, is as follows. Kremlin dis- tributes a Russian-language message on Twitter and other social media as part of a broader campaign, in more than one language, in the hopes that the topic will be "picked up" for discussion and make it to low-quality foreign news media, like Alex Jones’s Infowars in the . The message, then, has the potential to reach higher level conversations among policy makers. Not to be disregarded, also, are voters of US-Russian and French- Russian heritage. Many of these citizens of the US and France still follow Russian media online and elsewhere, and engage in regular conversations with their families in Russia. Their votes could be of some importance to the eventual outcome of the elections.

Background

Since the Arab spring, political scientists have been increasingly evaluating the effect that the Internet, especially social media, can have on beliefs and political behavior. Al- though a consensus on the effect of social media on political outcomes is still lacking, there is considerable evidence that social media might catalyze political action, such as protests. Benford and Snow (2000) find that social media might provide “frames" for a protest movement that would help propel this movement from small - scale to large - scale.7 Social media (and the Internet, more generally) have also been found to exacerbate social cleavages and worsen political polarization, due to a process termed “homophilous sorting." In other words, when people have relative control over information they are ex- posed to (for instance, seeing ideologically congruent news and posts on social media,

7Robert Benford and David Snow, “Framing processes and social movements: an overview and an assessment". Annual Review of Sociology 26, pp. 611 - 639, 2000.

5 as opposed to turning on the TV to be exposed to a stream of messaging, only some of which might be congruent), they will tend to solidify their political opinions and drift far- ther away from the center of the ideological spectrum.8 But although this chain of events might be detrimental to a democracy in a broader sense (by limiting its citizens’ exposure to alternative points of view and making consensus building more difficult), some pre- liminary evidence shows that “hashtag activism" can, via the same process, translate into a shared political temporality that will take political action from the online realm to the physical world.9 Social media also lower costs of collective action, for instance by replacing the re- quirement of centrally coordinating a protest with much more relaxed organizational mechanisms.10 This research has been applied to various regions. Stepanova (2009) describes the role that social media played in the Arab spring, and cautions that “No re- gion, state, or form of government can remain immune to the impact of new information and communication technologies on social and political movements."11 White and McAl- ister (2014) examine protests in Russia in 2011 - 2013, and find that some, but not all, social media (Facebook, in particular) had an effect on opinion about protests. 12This

8Sunstein, Cass R., #Republic: Divided Democracy in the Age of Social Media, Princeton, NJ: Press, 2018. 9Bonilla, Yarimar, “#Ferguson: Digital protest, hashtag ethnograhy, and the racial politics of social media in the United States," American Ethnologist 42 (1), 2015, pp. 4 - 17; Carney, Nikita, “All Lives Matter, but So Does Race: Black Lives Matter and the Evolving Role of Social Media," Humanity and Society 40(2), 2016, pp. 180 - 199. 10Clay Shirky, Here Comes Everybody: The Power of Organizing Without Organization, New York: Penguin, 2009; Clay Shirky, “The political power of social media," Foreign Affairs. 90, pp. 28âA¸S41,˘ 2011. 11Elena Stepanova, “The Role of Information Communication Technologies in the ’Arab Spring’", PONARS Eurasia Policy Memo No. 159. Available from: http://www.ponarseurasia.com. Last accessed 8 May, 2017. Other research on the role of social media in the Arab Spring includes Daniel Ritter and Alexander Trechsel, “Revolutionary Cells: On the Role of Texts, Tweets and Status Updates in Nonviolent Revolutions," Florence, Italy: European University Institute, 2011; Marc Lynch, “Young brothers in cy- berspace," Middle East Report, 245: 26, 2007; Marc Lynch, “After Egypt: the limits and promise of online challenges to the authoritarian Arab state," Perspectives on Politics, 9, pp. 301-10, 2011; and many other studies. 12Stephen White, and Ian McAllister, “Did Russia (Nearly) Have a Facebook Revolution in 2011? Social

6 might be indicative of the participants’ orientation toward the West, but does not change the underlying fact that organizers of the protests used social media to collect money for meals, a stage and equipment, distribute messages about the timing and the substance of the protests, disseminate videos demonstrating abuses of official power and satirized offi- cials, among other things.13 Similarly, an uptick in the US protest activity, that developed, in part, in response to the Trump presidency, has been fueled and facilitated by social me- dia. Protests like Black Lives Matter, the March for Our Lives, and the Women’s March all relied on social media to improve organizing activity.14 Governments, especially authoritarian ones, manage the threat they perceive emanat- ing from social media in different ways. Some, like Egypt and Tunisia during the Arab Spring, use crude methods like shutting down the Internet (this was proven unsuccess- ful). Belarus, unusually even for authoritarian governments, has occasionally used social media against its users to tracks down individuals who have used it to speak out against the government, and interrogate or jail them. A somewhat sophisticated variation on the first method is China’s practice, documented by King, Pan and Roberts (2013) that allows posts that are critical of the regime but shuts down those that call for any kind of collective action, no matter how benign.15 A newer method of authoritarian governments to deal with oppositional messages and to promote their own is the increasing appropriation of these online spaces to pro- mote their own narrative, or spread propaganda. One variant on this is China’s method of diverting attention from sensitive political topics. Un updated work by King, Pan and Roberts (2017), for example, shows how China’s so-called “50 cent army", the Chinese

Media’s Challenge to Authoritarianism," Politics 34:1, pp. 72 - 84, 2014. 13Northam, Jackie, “Russian Activists Turn to Social Media," NPR 13 January, 2012. 14Farhi, Paul, “How Mainstream Media Missed the March That Social Media Turned Into a Phe- nomenon," , 22 January, 2017. 15Gary King, Jennifer Pan, Margaret Roberts, “How Censorship in China Allows Government Criticism but Silences Collective Expression," American Political Science Review, 2: 107, pp. 1 - 18, 2013.

7 counterpart to Russia’s troll farms, inundates social media with bland, benign, positive posts, especially at potentially politically sensitive time, instead of promoting the govern- ment’s political message.16 Thus, the Chinese government diverts attention away from sensitive topics, seeking to set the agenda of the conversation, rather than persuade the populace on politically sensitive topics. Lacking a database of “leaked" tweets similar to the one used by King et al, I am showing a different story taking place in Russia. Russia’s Kremlin-funded news outlets, like Russia Today, as well as trolls who retweet the same RT article over and over again, do not shy away from political debate. Instead, they participate in political conversa- tions aggressively, seeking to intimidate, discourage and silence the opposition’s voice. Whereas only 5 years ago, Twitter’s political conversation belonged to the opposition,17 this has changed, with Kremlin’s message now dominating in the political conversation.

Russia’s Official Position on the US and French Election

Aside from numerous stories in Russia Today and similar outlets, the Kremlin’s official position toward the candidates can be surmised from speeches, statements and elites’ actions. Putin’s dislike for Hillary Clinton is well-documented. In 2008, responding to Clinton, who earlier said that he had “no soul", Putin quipped, “At a minimum, a head of state should have a head."18 This response was widely interpreted as a crude, genitalia-related joke that would tie in neatly with Putin’s well documented misogynist attitudes.19 Competently documented

16Gary King, Jennifer Pan, and Margaret Roberts, “How the Chinese Government Fabricates Social Media Posts for Strategic Distraction, not Engaged Argument," American Political Science Review, 2017. 17Bruce Etling et al, “Mapping Russian Twitter," Berkman Klein Center, 20 March, 2012. 18“Things You Didn’t Know about Russian President ," News.com.au. http://www.news.com.au/world/europe/things-you-didnt-know-about-russian-president-vladimir- putin/story-fnh81p7g-1226845669588, 2014. Last accessed 30 March, 2018. 19Ibid.

8 by Valerie Sperling (2015), misogyny, as well as machismo, homophobia and other re- gressive sex and gender attitudes, are used in contemporary Russia by the Kremlin and the oppositional forces alike as part of their legitimation strategy.20 This works in Russia because of a widespread acceptance of gender stereotypes - that is, reducing a woman, even a high level politician in a powerful hegemonic country, to a lower status with the abovementioned remark, is one way to reassert Putin’s own power under patriarchy. But although this more general disdain of all women is a tempting explanation to follow, it does not neatly accommodate Putin’s support for the French right wing politi- cian, Marine Le Pen. Additional evidence suggests that Putin’s disregard for Clinton goes beyond general gender norms or his own legitimation of power, and is, indeed, quite personal. In 2011, when Clinton publicly criticized Russia’s elections that drew thousands of people to protest against Putin, he fumed, “She [Clinton] said they were dishonest and unfair. [...] We need to protect ourselves from this kind of intervention."21 Putin seemed to believe that Clinton, almost singlehandedly, was responsible for triggering the protests that threatened the regime. In fact, although Putin is frequently heard assigning blame to various groups (ethnic, political etc)22, it is quite rare for him to point the finger at an individual23. This suggests an unusual distaste for Hillary Clinton, as a specific, concrete person. In contrast, the relationship between Putin and Trump has been overtly positive, until

20Sperling, Valerie, “Sex, Poltics, and Putin," Oxford: Oxford University Press, 2015. 21Michael Crowley, Julia Ioffe, “Why Putin Hates Hillary," , 25 July, 2016, found at http://www.politico.com/story/2016/07/clinton-putin-226153. Last accessed 9 May, 2017. 22For instance, Putin blamed the Jews and other ethnic minorities for interfering with the 2016 US elec- tion, see Park, Madison, “Putin: Maybe Jews or Minorities Behind US Election Interference", CNN 12 March, 2018. He also blamed Trump’s political opposition for poor relations between the US and Russia, see, for instance, Roth, Andrew and David Filipov, “Putin Blames Trump’s Political Opponents for Poor US - Russian Relations," The Washington Post, 14 December, 2018. 23In fact, this is a strategy he frequently reserves for domestic foes.

9 recently. In 2015, Putin openly and lavishly praised the then unlikely presidential candi- date saying, “He’s a really brilliant and talented person, without any doubt. It’s not our job to judge his qualities, that’s a job for American voters, but he’s the absolute leader in the presidential race,"24 Trump reciprocated profusely, tweeting and pronouncing his own admiration for President Putin at every opportunity. Although there is some argu- ment with the translation of the adjective “brilliant" (“yarkii" in Russian, which can also be translated as “bright", or even “flamboyant"), the general sense of the approval of this candidate remains apparent and uncontested. The precise reasons for Trump’s popularity with the Kremlin are a matter of some speculation. Madeleine Albright’s interpretation is that Trump is a “useful idiot" for Rus- sia - an ineffectual, easy to manipulate leader.25 If a hegemon like the U.S. elects a bum- bling leader like Trump, this understanding goes, this will weaken the overall position of the U.S. on the world arena for at least four years, and maybe more. Under President Trump, powerful alliances that position themselves counter to Russia might fall apart, the U.S. economy might stagnate, and even the U.S. military would have an ineffectual Commander-in-Chief. This view certainly partially explains why Putin would like to see a President like Trump in the White House. The other part of the explanation is very straightforward: Trump signaled his own approval of the Kremlin and its policies early on. Considering the sheer might of the US economic and military resources, having a friendly President in the White House makes ample geopolitical sense and inuitive sense. Early signaling by Trump that he did not believe in sanctions against Russia alone would have made him an attractive candidate

24Alex Griswold, “Putin: Donald Trump is ‘Brilliant’; Russia Welcomes His Presidency," Medi- aite 17 December, 2015, found at http://www.mediaite.com/online/putin-donald-trump-is-brilliant-russia- welcomes-his-presidency. Last accessed 9 May, 2017. 25Brent Griffiths, “Albright: Trump Fits the Mold of Russia’s ’Useful Idiot," Politico 24 Octo- ber, 2016. Found at http://www.politico.com/story/2016/10/trump-russia-useful-idiot-madeleine-albright- 230238. Last accessed 9 May, 2017.

10 as opposed to Hillary Clinton who was prepared to maintain the policy of her immediate predecessor. The logic of choosing a candidate to support in the French election was roughly simi- lar. Marine Le Pen was not the first candidate supported by Russia. The first, unfortunate, choice was Francois Fillon, who, as it quickly became apparent, did not have much of a chance at victory. Fillon’s presidential campaign was riddled with scandal and accu- sations of corruption. Unexpectedly, in March, his campaign received some help when a fake news report was published under the headline “The Return of Fillon to the Head of Opinion Polls."26 The origin of the story was found to lie with Sputnik, a Russian- government controlled news outlet. As early as 2011, Putin spoke favorably of the trend toward “conservative values" that we were observing in Europe, including the success of Marine Le Pen in municipal elections (she came in third).27 One signal that Putin was backing Le Pen came in March of 2017 when the President of Russia invited Marine Le Pen to Moscow. Anticipating the possible international reactions to the meeting with the leader of the National Front, Putin noted, “I am, of course, aware of the ongoing election campaign in France. In no case do we want to influence the events as they unfold, but we reserve the right to communicate with all representatives of all political forces in France, just like our partners in Europe and the United States are doing."28 Putin and Le Pen spoke at length about common security concerns - an indication of a plan toward a future cooperative relationship.

26Andrew Higgins, “It’s France’s Turn to Worry About Election Meddling by Russia," The New York Times, 17 April, 2017. 27“Direct Line with Vladimir Putin," transcript of a speech, found at http://en.special.kremlin.ru/events/president/news/20796. Last accessed 9 May, 2017. 28“Meeting with Marine Le Pen," trasncript of the meeting, found at http://en.kremlin.ru/events/president/news/54102. Last accessed 9 May, 2017.

11 Methods

Using Twitter’s API, I collected, in real time, a database of tweets, using select hashtags (for example, “ElectionsUSA", “Trump," “Clinton", “ElectionsFrance", and others29), for several days before, during, and after the US and the French election. The US election has yielded a total of 14,116 uniquely identified tweets; the French election, which I observed for a shorter period of time, yielded 2,917 tweets. Excluding retweets, each election has produced 5,127 tweets from 2,163 users; and 1,581 from 669 users respectively. This is not to say that retweets are unimportant. Metaxas et al (2015) show that retweets, or messages that were originated by another user and re-posted, verbatim, by another, indicate trust in the original source, and an agreement with the re-posted mes- sage.30 Twitter has introduced new support for retweets on their API31, which makes it much easier to find the number of retweets. Importantly, this also restricts the sample of retweets on the conservative side. Since the “retweet" button on Twitter does not al- low users to edit the content of the retweet, this prevents overcounting messages that use some of the language of the original post, but might edit the content to express disagree- ment. As I show below, a large number of retweets correlates strongly with the message aligning with the government-promoted narrative. Many of the retweets repeat a message produced and promoted by Kremlin - funded Russia Today. All data were collected in Python and analyzed R. I coded the content of the tweets manually, but quite conservatively (ending up, for that reason, with a great many tweets in the neutral/unclear category). If the message was strongly pro-Marine Le Pen or against Macron; or strongly pro-Trump and against Clinton, this was coded as friendly or un-

29These hashtags were all searched for in Russian. Full list is in the Appendix. 30Metaxas, P.T., E. Mustafaraj, K. Wong, L. Zeng, M. O’Keefe, S. Finn, “What Do Retweets Indicate? Results from User Survey and Meta-Review of Research," Proceedings of the Ninth International AAAI Conference on Web and Social Media, Wellesley College, 2015. 31In the form of a new, “retweet status", field.

12 friendly to the Kremlin’s message. Examples of Kremlin-friendly messages appear be- low.

Figure 1: Kremlin-friendly tweet 1.

Figure 2: Kremlin-friendly tweet 2.

The first tweet, which reads "Macron has promised to make Putin respect himself. Putin has promised to treat his ill imagination with understanding," paints Macron as a lesser figure to more “senior" Putin. In this tweet, Macron is both vying for Putin’s re- spect, and is delusional if he thinks he can earn it. This is clearly disparaging to Macron’s candidacy. The second tweet, which reads "Le Pen has called the a great Eu- ropean people," connects Le Pen’s candidacy to Russia’s geopolitical ambitions. This is clearly intended to promote her candidacy at home. Examples of Kremlin-unfriendly messages appear in Figures 3 and 4. The first Kremlin- unfriendly tweet reads, "Whatever you say, Trump, Hillary Clinton will be the President of the US." The second reads, "Mass media: Trump made donations to officials who in- fluenced his business." The first of these tweets is clearly an affirming one for the Clinton presidency; while the second one seeks to discredit Trump by pointing out his unethical dealings.

Figure 3: Kremlin-unfriendly tweet 1.

Figure 4: Kremlin-unfriendly tweet 2.

13 Automated text analysis, like counting up frequency of words, begins with the under- standing that the order of words does not matter when we are trying to understand the general meaning of a corpus of texts.32 Treating the corpus of tweets as a “bag of words" will allow the researcher a manageable set of data from which we can draw preliminary conclusions. Another important step that I rely on in the automated portion of this analy- sis is “stemming" - that is, reducing the overall complexity of words by reducing them to their roots. This is especially important for a grammatically complex language like Rus- sian. Thus, all the various forms of the name “Trump" (“Trampa", “Trampom", “Trampu" etc), that arise because of noun cases can be counted together as one word. Word clouds were produced using www.wordclouds.com. The most frequently ap- pearing terms (like Trump, Clinton etc, were stemmed for the purpose of the word clouds). All translations from Russian to English are the author’s.

Results and Discussion

We begin this analysis with the assumption that candidates’ performance on the Russian language Twitter will have a similar pattern to how the Russian state addressed both of the elections in English in the US, and in French for the French Twitter. We can make this assumption from the recent qualitative evidence on the Russian state’s concerted efforts to influence the elections in the US and France. 33 Sentiment analyses below also indicate a strong presence of the Kremlin-friendly narrative in the coverage of both elections. If the assumption holds, then these data are a useful first step toward understanding the strategy that Russia employed in influencing these elections.

32Grimmer, Justin, and Brandon M. Stewart, “Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts," Political Analysis 2013, pp. 1-31. 33See, for instance, Popken, Ben, “Twitter Deleted 200,000 Russian Troll Tweets. Read Them Here," NBC News 14 February, 2018, found at https://www.nbcnews.com/tech/social-media/now-available-more- 200-000-deleted-russian-troll-tweets-n844731. Last accessed 1 April, 2018.

14 But even if the assumption does not hold, and the patterns of discussing the two elec- tions are different domestically and internationally, understanding how Russian domestic audiences discussed the two elections provides an interesting snapshot of the Russian public opinion of both the elections and the newly elected leaders. In the absence of solid evidence that the Russian language tweets mirrored English and French-language tweets distributed abroad, the only remaining unclear point is whether the opinion of the Russian state mirrors that of the Russian public, or whether they diverge. The 2016 US election was talked about roughly the same amount as the 2017 French election. I started collecting content on the US election on 26 September, 2017, and col- lected, on average, 100 tweets per day. Data collection for the French election started later in the process - on 9 April, 2017, and continued until May 16th. It yielded, on average, a similar number of tweets per day. On the one hand, I would have expected France to be a more popular topic on Twitter, given that it is so much closer to Russia geographically. However, the United States has, for decades, been Russia’s great geopolitical rival, the “other" to define itself against. The United States is a frequent topic of coverage in the Russian news, both printed and televised. Both Kremlin, and the Russian public, have a lot of interest in the next president of the United States. During the US election, the two most popular words were, predictably, “Clinton" and “Trump" (see Figure 1 below). Note that to reduce noise, the graph above does not include words under 4 letters, or over 9 letters in length. The lower limit is in place in order to exclude words like prepositions that would amount to noise. The upper limit is in place to exclude typos that appear as a result of a missed space. It also excludes words that appear fewer than 500 times. However, this graph does not account for stemming. Once stemming is done (it had to be done by hand because of the difficulty automating some features of text analysis with

15 a language other than English), the word “Trump" clearly dominates over “Clinton" (see Figure 2 below). Other frequently mentioned words were “Putin," “debates", “Russia," “elections, “president," “huge," and, surprisingly, “monsters." This word, always in the plural, came up over 100 times in my sample of tweets. The content of these messages is, unfortunately, ambiguous (I coded them as neutral/unsure), and the meaning, although suggestive, remains unexplained.

Figure 5: Frequency of most common words, US election 2016 on Russian Twitter

16 Figure 6: Word Cloud, US election 2016 on Russian Twitter

The two graphs provide comparable visual statistics on the sheer popularity of cer- tain words in the tweets around this time. In the first case, even before accounting for stemming, Trump is clearly a more popular topic of conversation than Clinton. The word cloud, by graphically correlating the size of the word with its frequency of mention, brings this fact into even higher relief. That Donald Trump was talked about more is not insignificant. Even if the messages were negative, they would still provide free publicity for the candidate. Recent analyses within the US political context (Vaccari 2013) have shown that the volume of online coverage of candidates is a bigger determinant of a politician’s online popularity than such factors as candidates’ coverage in the mainstream media, performance in the opinion polls, or even how much money they spent in campaigns.34 Simply tweeting and re- tweeting Donald Trump’s and Marine Le Pen’s name in even neutral messages (a tweet like "On U.S. elections #Trump") increases the sense of this candidate’s presence and

34Vaccari, Cristian, “What Drives Politicians’ Online Popularity? An Analysis of the 2010 U.S. Midterm Elections," Journal of Information Technology and Politics 10(2), 2013, pp. 208 - 222.

17 inflates his perceived significance in the race. Reading the actual content of the tweets, however, provides an additional dimension to this observation. Trump is frequently portrayed in sympathetic tweets as the honest, masculine and trustworthy candidate, a worthy adversary to Clinton’s candidacy of status quo, bellicose relationship with Russia, and feminine weakness. The three tweets in fig- ures 7 - 9 below illustrate this point well. The first tweet reads, "Can it be that in order to elect Clinton, we need to start a world war? I’m irate." This taps into the oft-repeated narrative in the Russian contemporary discourse that the U.S. is a belligerent hegemon, seeking to either subjugate a country like Russia or start a war. The tweet in Figure 8, "Trump has accused Clinton in cheating on her husband," echoes the frequently used, although unsubstantiated accusation that Clinton lacked integrity (domestically, Trump used the same invective in an unflattering nickname, “Crooked Hillary," that he came up with during the U.S. campaign), and Trump, by accusing her, upholds it. The final, par- ticularly acrid, tweet reads, “How has she not fallen apart out of age, the evil old shrew? Clinton and the clinic." This is an example of a frequent ad hominem employed against women in power, one that goes after a superficial characteristic, like age or looks.

Figure 7: Pro-Trump/Anti-Clinton tweet.

Figure 8: Pro-Trump/Anti-Clinton tweet.

Figure 9: Pro-Trump/Anti-Clinton tweet.

Interestingly, the French election does not reveal a clear lead of Le Pen over Macron

18 in frequency of mention (those are roughly equal, with Le Pen slightly in the lead). This likely has to do with the late adoption by the Kremlin of Le Pen as the candidate of choice. It might also reflect the fact that the simple frequency of mentions of the can- didates includes both flattering, and unflattering messages. A common theme of anti- Macron messages (that did not even include Marine Le Pen) is that he is beholden to the global financial elites, too young, inexperienced etc. A common ad hominem attacked his choice of spouse who is considerably older than him. Another thing that stands out is the number of re-tweets in the French sample - there are almost as many re-tweets as the original messages. This gives further credence to the previous academic work that has shown how important re-tweets are to mapping sen- timent displayed on Twitter. Re-tweets work particularly well on Twitter because they require a minimal effort and cost, lower than producing original content. They also refer to a familiar source that becomes more familiar the more it is mentioned, in a circular, intensifying sort of a pattern. Re-tweets are also low risk and they allow “plausible deni- ability" of sorts. One might always claim that the tweet that was repurposed was either misunderstood or re-tweeted by accident, since the action required to repurpose another message is minimal. Another interesting feature of the French sample is the frequency of the word “Crimea" - many tweets relayed messages on different candidates’ positions on the annexed region. This suggests a continued salience of the matter, and a continued effort to justify the annexation. One such message, for example, produced by RT and re-tweeted multiple times was Francois Fillon’s statement on Crimea - that the peninsula belongs to Russia in a cultural and historic sense, and it is “useless" to continue to demand the Russians’ withdrawal from it. It is unclear what Fillon was trying to accomplish in France with a statement like this, but the Russian Twitter ran with the message, re-tweeting it dozens of

19 times.

Figure 10: Word Cloud, French Election 2017 on Russian Twitter

As a sentiment analysis of the presence of the two elections on the Russian Twitter, I manually coded the “tweets" as either Kremlin-friendly, Kremlin-unfriendly, or neu- tral/unsure. Because the US election data set is large (over 14,000 “tweets"), I code only a sample of 2,000. Although an automated process was not availale for this analysis, due to the language of the data set, I do not think that coder reliability is a concern here. The neutral/unsure category is so large because it catches a lot of ambiguous messages. Only those messages clearly expressing support for LePen, or Trump, or against their opponents were coded as “friendly" or “unfriendly." In both elections, the Kremlin-friendly message is clearly dominant. Much of this comes from RT, but many of the RT messages are picked up by individual, smaller users and rehashed. This provides some evidence to believe that Kremlin is appropriating the space of social media, to expand its array of platforms for propaganda. The final analysis I perform here is a simple OLS model to determine what kinds of factors correlate with support for Kremlin-promoted messages. For this, I recode the

20 sentiment variable as a scale, from -1 (message out of alignment with the Kremlin) to 1 (message aligning with the Kremlin). I find that mentioning RT anywhere in the sta- tus is a strong predictor of support for Kremlin-approved candidates, both in terms of robustness and in substantive terms. Specifically, mentioning RT in the status increases alignment with Kremlin’s message by 0.1 to 0.34. Number of re-tweets matters, which makes sense considering that many messages promoted by RT are intended to be dissem- inated as widely as possible. The number of followers that a user has plays no role in predicting their level of sup- port for the Kremlin-promoted message. This is also intuitive. Many trolls start new accounts and abandon old ones within a few days of creating them. These are not long- term users who worry about their screen name and their reputation online, but transitory, made-up personalities, with few, if any, friends or followers. The coefficient on the num- ber of statuses, although a statistically significant result in the US model, is substantively tiny. I disregard this here because of the small effect size.

21 Table 1: Determinants of support for Kremlin-promoted message on Twitter

Dependent variable sentiment (US) (France) Russia Today 0.104∗∗ 0.343∗∗∗ (0.042) (0.053)

Number of “re-tweets" 0.010∗∗∗ 0.007∗∗∗ (0.001) (0.002)

Number of followers 0.00000 0.00000 (0.00000) (0.00000)

Number of statuses 0.00000∗∗∗ 0.00000 (0.00000) (0.00000)

Constant 0.167∗∗∗ 0.097∗∗∗ (0.032) (0.035)

Observations 1,997 1,126 R2 0.057 0.103 Adjusted R2 0.055 0.100 Residual Std. Error 0.781 (df = 1992) 0.708 (df = 1121) F Statistic 30.231∗∗∗ (df = 4; 1992) 32.285∗∗∗ (df = 4; 1121) Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

22 Throughout this essay, I use the term ‘propaganda’ casually, although this belies the term’s complexity. The term can be used (albeit archaically) in a neutral way, to address things like the “propagation" of religious faith. In Lenin’s Russia, “propaganda" was also not a pejorative term. Rather, the distribution of Marxism and Leninism was seen as a service for the Communist state. The term’s most up-to-date usage, however is dismis- sive, pernicious, meant to convey both malicious intent beyond disseminating information (much of the time, although not necessarily, untrue), and a certain lack of sophistication on the part of the recipient, unable to discern that something is being sold to him or her. Some scholars of propaganda have pointed out that this latter use of the term is con- gruent with the development of new technologies, such as the radio, to disseminate infro- mation. Hilmes (2013) writes: “The “mass audiences" created by popular media seemed to constitute the very image of the propaganda-susceptible, emotional mob, faceless and unknowable, feared by the nations’ intellectual and political elites. On the other hand, the instruments of mass communication, especially radio, gave national elites the most

23 powerful tool yet available to them to attempt to shape and control those masses."35 This logic is once again at play here. The very aspects of social media that once made it so attractive to the opposition – its incredible reach, its flexibility and speed in creating and developing a message – are exactly what makes it so useful to Russia’s incumbent authoritarian regime. Furthermore, the international nature of some media, like Twitter, blur the line between national and international propaganda. Messages overtly intended for domestic audiences (because they are produced in Russian) might end up repeated in target countries by mass media outlets, citizens of foreign countries, and even policy makers. Like radio before it, Twitter transgresses geographical boundaries, and delivers chosen messages to users and minds.

Conclusion

Hybrid and authoritarian regimes rely on a variety of media to promote their message. Traditionally, the USSR, and then Russia under Putin, has relied on the newspaper and the TV to distribute propaganda. While there is no indication that the government no longer favors those methods, we now also see some indication of the Kremlin keeping up with the times by creating and promoting its own hashtags. Whereas previously, Russian propagandists concentrated on domestic matters and do- mestic platforms, this is no longer the case. Evidence from Russian Twitter has shown that the Kremlin’s presence on the platform is quite salient. Tweets supporting Putin-chosen candidates are plentiful and persistent. Many of them originate with and are propagated by Russia Today, a news outlet that was started years ago as a soft power tool for Russia, but has by now morphed into a full fledged vehicle for propaganda.

35Michele Hilmes, “’The New Vehicle of Nationalism’: Radio Goes to War," in Jonathan Auerbach and Russ Castronovo, eds., “The Oxford Handbook of Propaganda Studies," Oxford: Oxford University Press, 2013, pp. 201 – 202.

24 Further work will cross check messages from known Russian users of the American Twitter-sphere with messages propagated on the Russian language Twitter to see if the crucial assumption made in this paper holds.

Appendix

The complete list of the 2016 US election tweets (all searched for in Cyrillic):

1. “VyboryvSShA"

2. “VyborySShA"

3. “Trump"

4. “Clinton"

The complete list of the 2017 France election tweets (all search for in Cyrillic):

1. “VyboryFranciya"

2. “MarinLePen"

3. “Sarkozy"

4. “EmmanuelMacron"

5. “FransuaFiyon"

6. “Macron"

7. “LePen"

8. “Fiyon"

25