Battle for Britain: Analysing drivers of political tribalism in online discussions about Brexit Samantha North Lukasz Piwek Adam Joinson University of Bath University of Bath University of Bath Bath, UK Bath, UK Bath, UK s.north@bath.ac.uk lzp20@bath.ac.uk aj266@bath.ac.uk on the network effect of homophily, echo chambers in- crease polarization by diminishing the likelihood of ex- Abstract posure to conflicting viewpoints. This creates tribes, which in turn may have negative effects on social cohe- This position paper details ongoing work ex- sion and the health of democratic societies. Although ploring political tribalism in online discussions various studies have discussed online polarization, few about Brexit. We use computational methods have systematically explored the potential driving fac- to analyze a Twitter dataset of significant size tors of political tribalism from a computational stand- (over 7 million tweets spanning 32 months of point on a dataset of this size (over 7 million tweets conversations), using group identity keywords from 32 months of conversations). Possible driving fac- (e.g. Brexiteer, Remainer) as a proxy for trib- tors could include group conflict dynamics, automated alism. Initial results indicate that levels of amplification (e.g. by bots), reciprocity or disinforma- tribalism increase over time for all keywords, tion. We discuss ongoing work that uses computa- in particular for pro-EU ones (Remainer, Re- tional methods to analyze these factors in relation to moaner). We also find a number of anoma- the Brexit discussion on Twitter. lies in the volume of tribal keyword use over time, which may relate to real-life political 2 Related Work events. Here we discuss initial findings and briefly present ideas for further research. 2.1 Political Polarization Online Much research has established that social media en- 1 Introduction courages polarization of its users, according to the Virtual ’tribes’, or interest communities, have long principle of homophily[MSLC01]. Extreme group po- been common on the internet [AS08], but recent years larization is harmful to democracy and social co- have seen a distinct upswing in tribalism of a distinctly hesion because it risks diluting the environment of political nature. In the UK, tribalism has redefined free discussion that ought to characterize healthy the political landscape along new identity-based lines, democracies[Sun73]. People engulfed in echo chambers with many voters abandoning traditional voting pref- of their own making may be less capable of listening erences [HLT18]. Britain’s new tribes represent votes to or empathizing with the perspectives of others, es- in the 2016 EU referendum: Leave or Remain. The pecially those from the opposing political side. Other digital age has exacerbated political tribalism, in part studies have indicated that social media users do en- because social media users can easily cluster in echo gage in discussion, but do so in a way that reinforces chambers filled with like-minded individuals. Based rather than breaks down boundaries between groups [KFS05]. When discussion involves outgroup deroga- Copyright c 2019 for the individual papers by the papers’ au- tion, one-upmanship, and challenges to existing view- thors. Copying permitted for private and academic purposes. points, groups may risk becoming further polarized, This volume is published and copyrighted by its editors. known as the ’backfire effect’ [NR10]. A notable early In: A. Aker, D. Albakour, A. Barrón-Cedeño, S. Dori-Hacohen, M. Martinez, J. Stray, S. Tippmann (eds.): Proceedings of the study of US political blogs [AG05] demonstrated the NewsIR’19 Workshop at SIGIR, Paris, France, 25-July-2019, existence of online polarization, where bloggers from published at http://ceur-ws.org both sides of the political spectrum would primarily link to others on their own side. On Twitter, find- 3.2 Data Analysis ings from Yardi and Boyd further support this while To discover statistical anomalies for volume of key- also drawing an important link between online polar- words on any specific date, we ran the data through ization and group identity. Examining 30,000 tweets the R library anomalize. Next, we combined the events from users on both sides of the US abortion debate, timeline with the anomalies data to reveal relation- Yardi and Boyd found that group identity is strength- ships between events and anomalies. The three high- ened when like-minded users reply to each other. But est anomaly spikes related to the unprecedented par- when different-minded users reply, their group affili- liamentary defeat of Theresa May’s Brexit deal, the ations are reinforced [YB10]. More recent work pro- launch of the ’Chequers deal’ and the European Com- vides further support for this idea. Bail et al. found mission urging member states to prepare for a no-deal that groups of Twitter users (one Democrat, one Re- Brexit. publican) reinforced their existing views after repeated exposure to opposing ones [BAB+ 18]. 4 Initial Findings Initial results from the anomaly detection work show 2.2 Group Conflict and Threat Perception a general trend towards increased use of all keywords We turn to group conflict theories to further guide our over time, indicating an upswing in Brexit tribalism analysis. Previous work on intergroup behavior has online. This effect is more pronounced for the key- found evidence of tribal tendencies [SHW+ 54]; [TT79]. words ’Remainer’ and ’Remoaner’, which could indi- In particular, intergroup threat theory suggests that cate either outgroup derogation aimed at pro-EU vot- perceptions of threat (either realistic or symbolic) in- ers (Remoaner/Remainer), or pro-EU voters identi- crease the likelihood of two opposing groups behav- fying as their own tribe (Remainer). We have also ing negatively towards one another (outgroup deroga- found indications that anomalies increase around cer- tion), such as in political settings [ODD08]. Realistic tain Brexit-related events that could be viewed as sym- threat is defined as concern about physical harm or bolic threats. To reinforce these findings, our next step loss of resources, while symbolic threat is based on will be to conduct text analysis (likely using word2vec) concerns about challenges to ingroup identity and val- on tweets around each anomaly to understand whether ues [SYM10]. For Remainers, symbolic threat chal- perceptions of threat are driving them. Perceptions lenges a cosmopolitan, tolerant and open identity, of threat are not the only factor that may be driv- while for Leavers, threat is more likely to target ideas ing political polarization online; the existence of infor- of sovereignty, control and national pride. We iden- mation operations targeting Western democracies has tify Brexit tribalism through the presence of two pairs been much documented, and deliberate efforts may of keywords, Brexiteer and Remainer or Brextremist be taking place to manipulate the Brexit discussion and Remoaner; the second set more derogatory than [BM19] ; [HK16]. On Twitter, cyber armies often use the first. We hypothesize that the volume of these bots to amplify certain content. Seeding of disinfor- keywords will increase when real-life events occur that mation into the social media ecosystem is also com- either group may perceive as a symbolic threat. mon. Disinformation and tribalism are deeply linked, as one of the central goals of information operations is to divide Western societies over controversial politi- 3 Methods cal issues, such as abortion, immigration and national identity. In our follow up work, we will attempt to 3.1 Data Collection quantify the effects of both bot activity and disinform- Our dataset consists of tweets from June 1, 2016 to ing content on tribalism and political polarization in February 13, 2019, extracted from Twitters Historical the Brexit discussion online. PowerTrack API. We queried for two pairs of keywords (as described above) that we believed could be used to References indicate affiliation with a Brexit tribe and potentially [AG05] Lada A. Adamic and Natalie Glance. The negative attitudes towards the opposition. The raw political blogosphere and the 2004 US elec- JSON dataset was of significant size, so we extracted tion: Divided they blog. In Proceed- only the relevant columns for this analysis (text, date ings of the 3rd International Workshop on and tweet id). We then generated a frequency col- Link Discovery, LinkKDD ’05, pages 36– umn to count how many times each keyword occurred 43, New York, NY, USA, 2005. ACM. on each date. To construct our events timeline, we combined three existing Brexit timelines from British [AS08] Tyrone Adams and Stephen Smith. A tribe mainstream media sources. by any other name. In Tyrone Adams and Stephen Smith, editors, Electronic Tribes: Conflict and Cooperation: The Robbers A virtual world of geeks, gamers, shamans Cave Experiment. Toronto: York Univer- and scammers., chapter 1, pages 11–20. sity, 1954. University of Texas Press: Austin, 2008. [Sun73] Cass R. Sunstein. #Republic: divided [BAB+ 18] Christopher A. Bail, Lisa P. Argyle, Tay- democracy in the age of social media. lor W. Brown, John P. Bumpus, Haohan Princeton : Princeton University Press, Chen, M. B. Fallin Hunzaker, Jaemin Lee, 1973. Marcus Mann, Friedolin Merhout, and Alexander Volfovsky. Exposure to oppos- [SYM10] Walter G Stephan, Oscar Ybarra, and ing views on social media can increase po- Kimberly Rios Morrison. Intergroup threat litical polarization. Proceedings of the Na- theory. In T. Nelson, editor, Handbook of tional Academy of Sciences, 115(37):9216– Prejudice. Lawrence Erlbaum Associates, 9221, 2018. 2010. [BM19] Marco T. Bastos and Dan Mercea. The [TT79] Henri Tajfel and J.T. Turner. An inte- Brexit Botnet and user-generated hyper- grative theory of intergroup conflict. In partisan news. Social Science Computer G. Austin and S. Worchel, editors, The Review, 37(1):38–54, 2019. social psychology of intergroup relations, pages 33–37. Nova Science Publishers Inc., [HK16] Philip N. Howard and Bence Kollanyi. 1979. Bots, #Strongerin, and #Brexit: Com- putational propaganda during the UK-EU [YB10] Sarita Yardi and Danah Boyd. Dynamic referendum. Working Paper, Project on debates: An analysis of group polarization Computational Propaganda, 2016. over time on Twitter. Bulletin of Science, Technology and Society, 30:316327, 2010. [HLT18] Sara B. Hobolt, Thomas J. Leeper, and James Tilley. Divided by the vote: Af- fective polarization in the wake of Brexit. American Political Science Association, pages 1–34, 2018. [KFS05] J. Kelly, D. Fisher, and M Smith. Debate, division, and diversity: Political discourse networks in USENET newsgroups. Online Deliberation Conference 2005. Palo Alto: Stanford University, 2005. [MSLC01] Miller McPherson, Lynn Smith-Lovin, and James M Cook. Birds of a feather: Ho- mophily in social networks. Annual Review of Sociology, 27(1):415–444, 2001. [NR10] Brendan Nyhan and Jason Reifler. When corrections fail: The persistence of po- litical misperceptions. Political Behavior, 32(2):303–330, 2010. [ODD08] Danny Osborne, P.G. Davies, and A. Du- ran. The integrated threat theory and pol- itics: Explaining attitudes toward politi- cal parties. In Bettina P. Reimann, edi- tor, Personality and Social Psychology Re- search, chapter 2, pages 61–74. Nova Sci- ence Publishers Inc., 2008. [SHW+ 54] Muzafer Sherif, O.J. Harvey, B.J. White, W.R. Hood, and C. Sherif. Intergroup