The baking agents of discussion? The role of anger and compassion in the dynamics of conflictual online discussions in three countries
Svetlana Bodrunova, Ivan Blekanov, Nina Zhuravleva, Anna SmoliarovaBACKGROUND. The spread of affective content on social media, as well as user grouping based on affect (Papacharissi 2015), has been a focus of scholarly attention for over a decade. But, despite this, we lack evidence on what roles various particular emotions play in the dynamics of discussions on social media. Emotional contagion theory (Hatfield et al. 2014) adapted for social media suggests that diffusion of emotions happens on individual level and spreads virally. Other theories, like those of social influence or social learning (Young 2009), though, suggest multiple, hierarchical, and/or topically-restricted contacts. The idea of affective agenda (Coleman & Wu 2010) implies that the dynamics of an emotional discussion needs to be assessed on the aggregate level, and emotional fragments of the discussions may foster discussion growth and/or fragmentation. But the question remains: what role the emotions taken on aggregate level play in the discussion dynamics, being either catalyzers or inhibitors of the discussions, and what theory better fits to describe the role of emotional discourse in the discussion growth/ebbing. One may suggest that emotions of different stance (positive/negative) may spur/slow down the discussions in various ways. OBJECTIVES. We analyze the spread of two polar emotions – anger and compassion – in three Twitter discussions on inter-ethnic conflicts, namely Ferguson protests (the USA, 2014), the ‘Charlie Hebdo’ massacre (France, 2015), and mass harassment in Cologne (Germany, 2015-2016). By analyzing the co-dynamics of the overall discussions and these two emotions we can conclude whether the pattern of the spread of emotions and its link with the discussion dynamics is the same in various language segments of Twitter. DATA COLLECTION AND METHODS. The data we use were collected by our patented Twitter crawler in the aftermath of the conflicts and include altogether over 2,5M tweets. We used manual coding by native speakers and machine learning to detect the emotions, with the precision and accuracy levels no lower than 0,75 for all the three languages. Then, we visualized the dynamics of growth of the emotional content of the discussions and used Granger test to see whether anger or compassion gave a spur to the discussions. RESULTS. We have received moderate results in terms of the dependence of the number of neutral users upon that of emotional users, but have spotted that the beginnings of the discussions, as well as the discussion outbursts, depend more on compassionate, not on angry users, which needs more exploration. We have also shown that the hourly dynamics of emotions replicates that of the larger discussion, and the numbers of angry and compassionate users per hour highly correlate in all the cases. Thus, we support the idea of emotional agendas, rather than that of emotional contagion, to better describe the emotions-based dynamics of conflictual networked discussions.