The final speaker in this ECREA 2018 session is Svetlana Bodrunova, whose focus is on polarisation in Twitter-based discussions of inter-ethnic conflicts in the U.S., Germany, and Russia. She also notes that the debate about whether echo chambers and filter bubbles are real is still ongoing, and that attitudes towards political actors have been most researched to date; divergence in such attitudes is often interpreted as polarisation, but this often mistakes the formation of homophilous clusters for actual polarisation. Importantly, too, cluster formation is often non-binary, and instead leads to the development of multiple, overlapping, and dynamic thematic clusters.
In such analyses, the political views of participants are often assessed via proxies: by friendship affiliations, retweeting patterns, following patterns, content-sharing patterns, and user self-descriptions, but not so often by analysing the actual content of the messages posted by users. Sentiment analysis should be able to help here, in principle, but the problem in this case is the reliable assessment of positive or negative sentiment in user-selected language that may also be sarcastic or move between different registers and lexicons of language.
The present study examined how historical studies coded political language, therefore, and attempted to translate this to current contexts. It did so in the context of a Russian conflict between locals and migrants in Biryuliovo, in the Ferguson riots in the U.S., and the New Year’s Day controversy in Cologne. It attempted to identify similarly biased groups in the Twitter discussion in each case, and to assess the community structures that emerged from this.
The data used began with the most influential users in each case (using a number of posting and network metrics, for datasets captured using hashtags as well as keywords), and discovered that the sets of political actors discussed in each case were the same: they were federal leaders, regional authorities and police, nationalists, and ethnic minorities. Influencers were clustered by multi-dimensional coding, and the project identified the vocabularies of the most divergent clusters.
In Russia, there was a neutral media discourse cluster, two different nationalist clusters (both against state power, and in favour of state power), and a cluster of neutral users; in Germany, there were pro- and anti-immigrant clusters as well as an overlapping more neutral group; in the U.S., there were media-oriented users, people blaming white supremacists, nationalists, and various other groups. A simple binary, left-right representation of polarisation would therefore be highly simplistic and misleading – and the ‘echo chamber’ model of polarisation is also insufficient. Echo chambers may exist at the extreme of the discussion, but at the centre there is considerable cross-interaction. Emotions appear to be key in driving the dynamics of such clusters.