Online Emotions and Offline Political Violence: Insights from the Far Left

With their exponential growth, the internet and online social networks are playing an increasingly significant role in political conflicts and contentious situations, as politicized online environments provide a growing potential for outbreaks of political violence. In the post-2008 context of polycrisis, the rise of online political networks has coincided with a resurgence of political violence in Europe and the United States. Both individual and collective emotions are closely associated with political violence; yet the ways in which online emotions are linked to real-life violence remain underexplored. This paper seeks to advance understanding of the emotional foundations underlying the connection between extremist online discourse and offline violent behavior. It achieves this by analyzing two novel datasets that systematically capture data on extremist discussions and actions within far-left communities during the highly significant period of the financial and refugee crises in Greece. Drawing from political psychology theories, computer science methods and statistical analysis, we assess and analyze emotions in a large corpus of far left discussions and investigate their association with real-life far left violence. More specifically, we conduct an emotion analysis of 308.327 online articles from the Greek far left political spectrum and examine the association of emotional political discourse with offline incidents of far left political violence over an 11-year period, spanning from 2008 to 2018. We begin by massively translating five well-known emotion datasets from English into Greek to develop a model capable of predicting emotions in a corpus of far left articles. Using six basic emotions—anger, fear, sadness, neutral, surprise, and happiness—we explore the relationship between online emotional expressions and an offline dataset of recorded far left violence. The research results indicate that there is such a connection of online emotions with offline recorded violence. Using a vector autoregression (VAR) analysis on the incidence of violence and different measures of intensity for various emotions, we find that although emotions cannot predict low escalation incidents of violence, that is low intensity violence that presents no threat of harm for human life or integrity, incidents of high escalation are associated and can be predicted by variation in anger in daily discourse. In particular, when the intensity of anger in the right tail of the daily anger distribution rises, we expect an increase in the incidents of high escalation of violence after a day. Reversely, emotions cannot be predicted by other emotions or violence, possibly indicating a direction of causation from emotions to violence but not vice versa. From a theoretical standpoint, our results can help analyse the emotional mechanisms through which acts of violence are triggered. From a policy perspective, our results can help predict periods of increased risk of high intensity violence at a granular (daily).