The Politics of Social Media Manipulation The Politics of Social Media Manipulation Edited by Richard Rogers and Sabine Niederer Amsterdam University Press Cover design: Coördesign, Leiden Typesetting: Crius Group, Hulshout isbn 978 94 6372 483 8 e-isbn 978 90 4855 167 5 (pdf) doi 10.5117/9789463724838 nur 670 Creative Commons License CC BY NC ND (http://creativecommons.org/licenses/by-nc-nd/3.0) All authors / Amsterdam University Press B.V., Amsterdam 2020 Some rights reserved. Without limiting the rights under copyright reserved above, any part of this book may be reproduced, stored in or introduced into a retrieval system, or transmitted, in any form or by any means (electronic, mechanical, photocopying, recording or otherwise). Table of Contents 1 The politics of social media manipulation 19 Richard Rogers and Sabine Niederer 2 Political news on Facebook during the 2019 Dutch elections 71 Stijn Peeters and Richard Rogers 3 Political news in search engines 97 Exploring Google’s susceptibility to hyperpartisan sources during the Dutch elections Guillén Torres and Richard Rogers 4 The circulation of political news on Twitter during the Dutch elections 123 Sabine Niederer and Maarten Groen 5 Dutch political Instagram 147 Junk news, follower ecologies and artificial amplification Gabriele Colombo and Carlo De Gaetano 6 Dutch junk news on Reddit and 4chan/pol 169 Sal Hagen and Emilija Jokubauskaitė 7 Fake news and the Dutch YouTube political debate space 217 Marc Tuters 8 Conclusions 239 Mainstream under fire Richard Rogers and Sabine Niederer 9 Epilogue 253 After the tweet storm Richard Rogers and Sal Hagen References 257 Index 287 List of figures and tables Figures Figure 1.1 Cartoon that ridicules the fake news taskforce, stating: “internet trolls are best countered by internet hobbits” 37 Source: Reid et al. (2018) Figure 1.2 “Detected and eliminated” fake news, with a warning issued by NU.nl and Nieuwscheckers 38 Source: NOS (2017a) Figure 1.3 The birth of the fake news crisis, or ‘fake news’ outperforms ‘mainstream news’ on Facebook, in the run-up to the U.S. elections in 2016 43 Source: Silverman (2016) Figure 1.4 Facebook political ad library tool, results for Britain’s Future, 13 March 2019 50 Figure 2.1 Engagement of mainstream (blue) and junk-like news (pink) articles found through provincial elections- related BuzzSumo queries, per week, between 18 February 2019 and 25 March 2019. Engagement scores have been normalised. 80 Line graph; visualisation by Federica Bardelli Figure 2.2 Total Facebook Engagement of fake versus mainstream news. Results from election-related queries on BuzzSu- mo, for the 20 most-engaged with articles between February and November 2016, per three-month period 81 Source: Silverman (2016) Figure 2.3 Per-query engagement of mainstream (blue) and junk (pink) articles found through provincial elections- related BuzzSumo queries, per week, between 18 February and 25 March 2019. Engagement scores have been normalised. 82 Line graphs; visualisation by Federica Bardelli Figure 2.4 Engagement of mainstream and junk-like articles found through EU elections-related queries on BuzzSumo, between 19 April 2019 and 23 May 2019. Engagement scores have been normalised. 84 Line graph; visualisation by Federica Bardelli Figure 2.5 Per-query engagement of mainstream (blue) and junk (pink) articles found through EU parliamentary election-related BuzzSumo queries, per week, between 19 April 2019 and 23 May 2019. Engagement scores have been normalised. 85 Line graphs; visualisation by Federica Bardelli Figure 2.6 Engagement of mainstream, hyperpartisan, conspiracy and clickbait articles found for provincial elections-related queries on BuzzSumo, between 18 February 2019 and 25 March 2019. Engagement scores have been normalised. GeenStijl is considered ‘mainstream’ here, while The Post Online is classified as ‘hyperpartisan’. 88 Line graph; visualisation by Federica Bardelli Figure 2.7 Engagement of mainstream, tendentious, hyper- partisan, conspiracy and clickbait articles found for provincial elections-related queries on BuzzSumo, between 18 February 2019 and 25 March 2019. Engage- ment scores have been normalised. GeenStijl and The Post Online are considered ‘tendentious’ here. 88 Line graph; visualisation by Federica Bardelli Figure 2.8 Engagement of mainstream, tendentious, hyperpar- tisan, conspiracy and clickbait articles found for EU parliamentary elections-related queries on BuzzSumo, between 19 April 2019 and 23 May 2019. Engagement scores have been normalised. GeenStijl is considered ‘mainstream’ here while The Post Online is classified as ‘hyperpartisan’. 89 Line graph; visualisation by Federica Bardelli Figure 2.9 Engagement of mainstream, tendentious, hyperpar- tisan, conspiracy and clickbait articles found for EU parliamentary elections-related queries on BuzzSumo, between 19 April 2019 and 23 May 2019. Engagement scores have been normalised. GeenStijl and The Post Online are considered ‘tendentious’ here. 89 Line graph; visualisation by Federica Bardelli Figure 2.10 Relative engagement of content categories across 4chan /pol/, Reddit, Twitter and Facebook. GeenStijl is considered ‘mainstream’ here while The Post Online is classified as ‘hyperpartisan’. 92 4chan and reddit data from 1 Dec 2015 until 1 June; Twitter and Facebook data from 18 Feb 2019-25 Mar 2019 and 19 Apr 2019-23 May 2019 Figure 2.11 Relative engagement of content categories across 4chan /pol/, Reddit, Twitter and Facebook. 92 4chan and reddit data from 1 Dec 2015 until 1 June; Twitter and Facebook data from 18 Feb 2019-25 Mar 2019 and 19 Apr 2019-23 May 2019 Figure 3.1 Presence of junk news in Google.nl search engine results for political queries related to foreign affairs, 13-22 March 107 Figure 3.2 Presence of junk news in Google.nl search engine results for political queries related to polarizing topics, 13-22 March 107 Figure 3.3 Presence of junk news in Google.nl search engine results for political queries related to the environment, 13-22 March 109 Figure 3.4 Presence of junk news in Google.nl search engine results for political queries related to the economy, 13-22 March 113 Figure 3.5 Presence of junk news in Google.nl search engine results for political queries related to societal issues, 13-22 March 112 Figure 3.6 Presence of junk news in Google.nl search engine results for political queries related to future innova- tion, 13-22 March 2019 112 Figure 3.7 Presence of junk news in Google.nl search engine results for political queries related to the environment, using language from the Facebook comment space of the political parties, 13-22 March 2019 113 Figure 3.8 Presence of junk news in Google.nl search engine results for political queries related to foreign affairs, using language from the Facebook comment space of the political parties, 13-22 March 2019 114 Figure 3.9 Presence of junk news in Google.nl search engine results for political queries related to polarizing topics, using language from the Facebook comment space of the political parties, 13-22 March 2019 114 Figure 3.10 Presence of junk news in Google.nl search engine results for political queries related to migration, using language from the Facebook comment space of the political parties, 13-22 March 2019 115 Figure 3.11 Presence of junk news in Google.nl search engine results for political queries related to migration and European Union issues, 22-24 May 2019 117 Figure 3.12 Presence of junk news in Google.nl search engine results for political queries related to climate and economic issues, 22-24 May 2019 117 Figure 4.1 Political party leaders as trolling targets on Twitter during the 2017 Dutch general elections. Each dot represents one mention (by a user mentioning political leaders at least 100 times). Red represents an attack, green represents a favourable mention. 126 Source: Borra et al., 2017 Figure 4.2 Engagement of mainstream (blue) and junk news (pink) articles during the Dutch Provincial election campaign (left) and the European Election campaign period (right) 129 Line graphs; visualisation by Federica Bardelli Figure 4.3 Engagement with mainstream news (blue) and junk news (pink) for the issue of MH17 (top right) and Zwarte Piet (top left) during the Provincial elections, and the EU elections (bottom right and left) 130 Line graphs; visualisation by Federica Bardelli Figure 4.4 Tweet and user counts, top hashtags, and most- retweeted tweets during the Dutch provincial election period of 2019 134 Dashboard; visualisation by Carlo De Gaetano Figure 4.5 Gephi visualisation of Zwarte Piet host-user network during the provincial elections campaign period, depicting only junk and tendentious hosts and the user accounts that circulate these sources 136 Visualisation by Carlo De Gaetano Figure 4.6 Gephi visualisation of MH17 host-user network during the provincial elections campaign period, depicting only junk and tendentious hosts and the user accounts that circulate these sources 137 Visualisation by Carlo De Gaetano Figure 4.7 Gephi visualisation of Utrecht shooting host-user network during the provincial elections campaign period, depicting only junk and tendentious hosts and the user accounts that circulate these sources 138 Visualisation by Carlo De Gaetano Figure 4.8 Gephi visualisation of PS2019 host-user network during the provincial elections campaign period, depicting only junk and tendentious hosts and the users that circulate these sources 139 Visualisation by Carlo De Gaetano Figure 4.9 Gephi visualisation of Party Leadership host-user network during the provincial elections campaign period, depicting only junk and tendentious hosts and the users that circulate these sources 140 Visualisation by Carlo De Gaetano These line graphs visualise the engagement with mainstream news (blue) and junk news sources (pink) during the Dutch provincial election campaign (PS) and the European Election campaign period (EU), similar to Figure 2, but excluding the tendentious- hyperpartisan sources. 143 Visualisation by Federica Bardelli These line graphs visualise the engagement with mainstream news (blue) and junk news sources (pink) for the issues of MH17 and Zwarte Piet during the provincial elections (PS), and the EU elections (EU), similar to Figure 3, but excluding the tendentious- hyperpartisan sources. 130 Visualisations by Federica Bardelli Figure 5.1 Diagram of the research protocol, showing the type of hashtags and accounts used for querying Instagram, and the tools used to collect, visualize and analyze the data 151 Figure 5.2 Proportions of most liked content shared around the 2019 Dutch provincial elections, categorised as fake, satire, and not fake 154 Data source: Instagram Scraper; data collection: 25-28 March 2019; pie charts Figure 5.3 20 most liked posts per hashtag shared around the 2019 Dutch provincial elections, sorted from right (most junk) to left (least junk) 155 Data source: Instagram Scraper; data collection: 25-28 March; image wall Figure 5.4 Examples of the posts flagged as hyperpartisan or satire Data source: Instagram Scraper; data collection: 25-28 March; image wall 156 Alternate figure 4.2 Alternate figure 4.3 Figure 5.5 Proportions of most liked content shared around the 2019 European elections, categorised as junk and not junk 157 Data source: Instagram Scraper; data collection: 22 May 2019; pie charts Figure 5.6 20 most liked posts per hashtag shared around the 2019 European elections, sorted from right (most junk) to left (least junk) and grouped by type (elections, issues, political leaders, and parties). Posts flagged as hyperpartisan are colored in red. 158 Data source: Instagram Scraper; data collection: 22 May 2019; image wall Figure 5.7 Follower ecologies in the Dutch political space, visual- ized as a co-follower network and manually annotated. In the network, accounts with higher amounts of shared followers (pink) are placed closer to each other. 161 Data source: Phantombuster; data collection: 25- 28 March; network graph Figure 5.8 Degree of account fakeness according to report by the HypeAuditor tool. Accounts on the further right have more suspected ‘fake followers’ than accounts on the left side of the graphs. 162 Data source: HypeAuditor; data collection: 25-28 March 2019; bee swarm plot Figure 5.9 Visualization of the follower base of Mark Rutte’s personal and work accounts and Geert Wilders’ account, based on results from the HypeAuditor tool. Each follower base is segmented based on ‘audience type’ and geographical provenance. Popular suspicious countries, that may suggest an inauthentic follower base, are coloured in red. 163 Data source: HypeAuditor; data collection: 25-28 March 2019; pie charts Figure 6.1 The frontpage of Reddit (retrieved 11-Jun-2019) 175 Figure 6.2 The index page of 4chan/pol/ (retrieved 11-Jun-2019) 175 Figure 6.3 Total amount of posts and comments on one of the Dutch subreddits (appendix I) 181 Data source: 4CAT and Pushshift; timeframe: 1-Dec- 2015 to 1-Jun-2019; line graph; visualisation by Gabriele Colombo Figure 6.4 Frequency of posts linking to Dutch junk news domains on Reddit 182 Data source: Google BigQuery; timeframe: 1-Dec-2015 to 1-Jun-2019; stream graph; visualisation by Gabriele Colombo Figure 6.5 Dutch versus non-Dutch subreddits in which Dutch junk news appears. Size of circle represents the overall number of posts in that subreddit within the timeframe, and colour represents the relative amount of posts with junk news. 183 Data source: Google BigQuery; timeframe: 1-Dec-2015 to 31-Jan-2019; circle pack diagram; visualisation by Gabriele Colombo Figure 6.6 Dutch subreddits where Dutch junk news appear compared to the size of all Dutch subreddits. Size of circle represents the overall number of posts in that subreddit, and colour represents the relative amount of posts with junk news. 184 Data source: Google BigQuery; timeframe: 1-Dec-2015 to 31-Jan-2019; circle pack diagram; visualisation by Gabriele Colombo All Dutch and non-Dutch subreddits where Dutch junk news appear compared to the size of all of Reddit. Size of circle represents the overall number of posts in that subreddit, and colour represents the relative amount of posts with junk news. 185 Data source: Google BigQuery; timeframe: 1-Dec-2015 to 31-Jan-2019; circle pack diagram; visualisation by Gabriele Colombo Figure 6.9 Line graph of posts with Dutch country flags on 4chan/pol/ 187 Data source: 4CAT; timeframe: 1-Dec-2015 to 01-Jun- 2019; line graph; visualisation by Gabriele Colombo Figure 6.10 Frequency of posts linking to Dutch junk news domains on 4chan/pol/ 188 Data source: 4CAT; timeframe: 1-Dec-2015 to 01-Jun- 2019; streamgraph; visualisation by Gabriele Colombo Figure 6.11 Links to news (red) and non-news (blue) sources in posts in Dutch subreddits 189 Figures 6.7 & 6.8 Data source: 4CAT and Pushshift; timeframe: from 1-Dec-2015 to 01-Jun-2019; treemap diagram; visualisa- tion by Gabriele Colombo Figure 6.12 Links to news (red) and non-news (blue) sources in Dutch posts on 4chan/pol/ 191 Data source: 4CAT; timeframe: 1-Dec-2015 to 1-Jun- 2019; treemap diagram; visualisation by Gabriele Colombo Figure 6.13 Links to Dutch (orange) and non-Dutch (blue) news on Dutch subreddits 191 Data source: 4CAT and Pushshift; timeframe: 1-Dec- 2015 to 01-Jun-2019; treemap diagram; visualisation by Gabriele Colombo Figure 6.14 Links to Dutch (orange) and non-Dutch (blue) news on Dutch subreddits 192 Data source: 4CAT; timeframe: from 1-Dec-2015 to 01-Jun- 2019; treemap diagram; visualisation by Gabriele Colombo Figure 6.15 Categories of news domains in posts on Dutch subreddits 193 Data source: 4CAT and Pushshift; timeframe: 1-Dec- 2015 to 01-Jun-2019; treemap diagram; visualisation by Gabriele Colombo Figure 6.16 Categorised types of news from news sources posted 4chan/pol/ 194 Data source: 4CAT; timeframe: 1-Dec-2015 to 1-Jun-2019; treemap diagram; visualisation by Gabriele Colombo Figure 6.17 Mean Reddit posts scores by Dutch junk news propa- gators (users who posted a link to a Dutch junk news domain at least twice) as reported by Pushshift API 197 Data source: 4CAT and Pushshift; timeframe: 1-Dec-2015 to 01-Jun-2019; bar graph; visualisation by Gabriele Colombo Figure 6.18 Subreddits where Dutch junk news domains are most often posted in 198 Data source: 4CAT and Pushshift; timeframe: 1-Dec- 2015 to 31-Jun-2019; circle pack diagram; visualisation by Gabriele Colombo Figure 6.19 Most linked to junk news domains on all of Reddit 199 Data source: 4CAT and Pushshift; timeframe: 1-Dec- 2015 to 1-Jun-2019; circle pack diagram; visualisation by Gabriele Colombo Figure 6.20 The top 1008 most posted YouTube videos in Dutch subreddits. Black labels denote deleted videos/chan- nels. Ranked left to right, top to bottom 204 Data source: 4CAT, Pushshift, and YouTube API; image wall Figure 6.21 The top 1008 most posted YouTube videos in Dutch subreddits, with video categories as an overlay. Black labels denote deleted videos/channels. Ranked left to right, top to bottom 204 Data source: 4CAT, Pushshift, and YouTube API; image wall Figure 6.22 The top 1008 most posted YouTube videos in 4chan/ pol/in posts with a Dutch country flag. Black labels denote deleted videos/channels. Ranked left to right, top to bottom 205 Data source: 4CAT and YouTube API; image wall Figure 6.23 The top 1008 most posted YouTube videos in 4chan/ pol/in posts with a Dutch country flag, with video categories as an overlay. Ranked left to right, top to bottom. Black labels denote deleted videos/channels 205 Data source: 4CAT and YouTube API; image wall Figure 7.1 Related channels on YouTube. Table where the top row displays the name of each Dutch political party and the columns below each of these are the media organizations associated with each party’s YouTube channel. 29 March 2019 222 Figure 7.2 TheLvkrijger post: Translated into English: “He who is silent agrees! Don’t shut up anymore! This is your country! Claim it!” 224 Figure 7.3 Related channels on YouTube. Panoramic graph of larger Dutch YouTube media sphere. This graph was reproduced two months apart on 29 March 2019 and again on 22 May 2019 with nearly identical outcomes. 225 Visualisation by Federica Bardelli using Gephi (Basian et al., 2009) Figure 7.4 Thumbnail diagram of the ‘fringe channels’’ top ten most popular videos 229 Visualisation by Federica Bardelli Figure 7.5 Screenshot from the “About” page on Cafe Weltschmertz’s YouTube channel which includes a sarcastic “trigger warning” for viewers whom might be angered by its frank approach to political debate, as well as crypto-normative espousal of “democratic hygiene processes”. 229 Figure 7.6 Weighted word lists of the titles of all the videos from the political commentary channels 231 Visualisation by Federica Bardelli Figure 7.7 Screenshot of a comment under the video of ‘Leukste YT Fragmenten’, referring to a ‘hopeless debate’ and the lack of consensus on the definition of ‘nepnieuws’ 232 Figure 7.8 Related channels on YouTube, 22 May 2019 237 Table where the top row displays the name of each Dutch political party who ran candidates in the EU election. As with figure 1, the columns below each of these are the media organizations associated with each party’s YouTube channel. The related channels for the parties are identical to figure 1 apart from a few minor differences and the fact that D66 now no longer returns any related channels, as with PvdA. Note also that of the two EU parties that return channels are categorized quite differently than the other national Dutch political parties. Tables Table 1.1 Overview of 2016 fake rallies planned and promoted, as listed in the US indictment of 13 Russian nationals concerning foreign election interference 26 Source: Parlapiano and Lee (2018) Table 2.1 Top 10 sites per category (provincial elections), for all queries combined, sorted by overall engagement scores as reported by BuzzSumo 87 Table 2.2 Top 10 sites per category (EU parliamentary elections), for all queries combined, sorted by overall engagement scores as reported by BuzzSumo 87 Table 2.3 Top 10 ‘hyperpartisan’ sites for both data sets (provin- cial and EU elections), sorted by overall engagement scores as reported by BuzzSumo 87 Table 3.1 List of Dutch political parties under study 103 Table 3.2 List of categories and political keywords used in the study 104 Table 4.1 Query overview showing the election campaign period (Provincial, EU or both), the political or issue space and the query made resulting in Twitter data sets 128 Table 5.1 Lists of hashtags pertaining to political leaders and politically charged discussions used to demarcate the Dutch political space on Instagram around the 2019 provincial elections 154 Table 5.2 Lists of hashtags pertaining to political leaders and politically charged discussions used to demarcate the Dutch political space on Instagram during the months before the 2019 European elections 157 Table 6.1 The top 3 best performing posts linking to a Dutch junk comain on Reddit 197 Data source: 4CAT and Pushshift; timeframe: 01-Dec- 2015 to 01-Jun-2019 Table 6.2 Metrics of users who shared the Dutch junk news on Reddit 200 Data source: 4CAT and Pushshift; timeframe: 01-Dec- 2015 to 01-Jun-2019 Table 6.3 The most occurring YouTube channels from all YouTube links posted in the Dutch Reddit and 4chan/pol/ samples 206 Data source: 4CAT, Pushshift, and YouTube API; timeframe: 01-Dec-2015 to 01-Jun-2019 Table 6.4 Compiled list of Dutch subreddits 211 Table 6.5 Junk news categorisation (expert list) Edited and enhanced list originating from Hoax-Wijzer 212 Table 6.6 Metrics for the proportions of news, Dutch news, Dutch junk news, and categories in posts on Dutch language subreddits, 01-Dec-2015 to 01-Jun-2019 214 Table 6.7 Metrics for the proportions of news, Dutch news, Dutch junk news, and categories in posts on 4chan/ pol/ with a country flag from the Netherlands, 01-Dec- 2015 to 01-Jun-2019 215 Table 6.8 Most occurring URLs from posts containing links to RT.com and Sputnik by posts with a Dutch country flag on 4chan/pol/ 215 Derived with 4CAT 1 The politics of social media manipulation Richard Rogers and Sabine Niederer Abstract This chapter gives an overview of the contemporary scholarship surround- ing ‘fake news’. It discusses how the term has been deployed politically as a barb against the free press when publishing inconvenient truths since the mid-nineteenth century. It also addresses how such notions have been used in reaction to novel publishing practices, including to the current social media platforms. More generally, the scholarship could be divided into waves, whereby the first related to the definitional issues and the production side, whilst the second has been concerned with its consumption, including the question of persuasion. There is additionally interest in solutions, including the critique of the idea that automation effectively addresses the problems. It concludes with research strategies for the study of the pervasiveness of problematic information across the internet. Keywords: fake news, junk news, disinformation, clickbait, hyperpartisan, post-truth Introduction: Influence campaigning in political spaces online and the question of persuasion In reviewing the scholarship surrounding so-called fake news, one would out of necessity make a distinction between the dominant work on the art of influence campaigning and computational propaganda online and the consequences to date for its consumers, but also the few findings, often journalistic, in the relatively understudied case of the Dutch political space online, both on the web as well as in social media. Much work has Rogers, Richard, and Sabine Niederer (eds), The Politics of Social Media Manipulation . Amsterdam, Amsterdam University Press 2020 doi: 10.5117/9789463724838_ch01