Us vs. Them: A Dataset of Populist Attitudes, News Bias and Emotions

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 1921–1945 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 1921 Us vs. Them: A Dataset of Populist Attitudes, News Bias and Emotions Pere-Llu´ıs Huguet Cabot1,3, David Abadi2, Agneta Fischer2, Ekaterina Shutova1 1 Institute for Logic, Language and Computation, University of Amsterdam 2 Department of Psychology, University of Amsterdam 3 Babelscape Srl, Sapienza University of Rome perelluis1993@gmail.com {d.r.abadi, A.H.Fischer, e.shutova}@uva.nl Abstract Computational modelling of political dis- course tasks has become an increasingly im- portant area of research in natural language processing. Populist rhetoric has risen across the political sphere in recent years; how- ever, computational approaches to it have been scarce due to its complex nature. In this paper, we present the new Us vs. Them dataset, con- sisting of 6861 Reddit comments annotated for populist attitudes and the first large-scale com- putational models of this phenomenon. We investigate the relationship between populist mindsets and social groups, as well as a range of emotions typically associated with these. We set a baseline for two tasks related to pop- ulist attitudes and present a set of multi-task learning models that leverage and demonstrate the importance of emotion and group identifi- cation as auxiliary tasks. 1 Introduction Political discourse is essential in shaping public opinion. The tasks related to modelling political rhetoric have thus been gaining interest in the natu- ral language processing (NLP) community. Many of them focused on automatically placing a piece of text on the left-to-right political spectrum. For instance, much research has been devoted to de- tecting bias in news sources (Kiesel et al., 2019) and predicting the political affiliation of politicians (Iyyer et al., 2014) and social media users, more generally (Conover et al., 2011; Pennacchiotti and Popescu, 2011; Preot¸iuc-Pietro et al., 2017). Other works conducted a more fine-grained analysis, iden- tifying the framing of political issues in news ar- ticles (Card et al., 2015; Ji and Smith, 2017). Re- cently, the field has also turned attention towards modelling the spread of political information in so- cial media, such as detecting fake news or political perspectives (Li and Goldwasser, 2019; Chandra et al., 2020; Nguyen et al., 2020). Populism has taken the spotlight in political communication in recent years. Various countries around the globe have experienced a surge of pop- ulist rhetoric (Inglehart and Norris, 2016) in both the public and political space. Despite this, ap- proaches to computational modelling of populist discourse have so far been scarce. Due to the flexi- ble nature of populism, annotating populist rhetoric in text is challenging, and the existing research in this area has relied on small-scale analysis by ex- perts (Hawkins et al., 2019). In this paper, we present a new dataset1 of Reddit comments anno- tated for populist attitudes and the first large-scale computational models of this phenomenon. We rely on research in social- and behavioural sciences (e.g., political science and social psychology) to operationalise a definition of populism and an …

Content truncated to 500 words