Talk like a populist, think like a populist? A text-based evaluation of populist discourse on social media

Few concepts in social science are as elusive as that of populism: is it an ideology, a rhetorical style, or just a winning strategy for elections? Due to its inherent elusiveness, populism has been assigned as many definitions as the forms it can take in the real world. This is a long-standing challenge for researchers of populism: how can we aim to study populism on a cross-cultural, multinational scale, if there is no consensus about what it actually is? The last decade has opened up new possibilities, particularly with respect to quantitative approaches to the social sciences: the widespread use of social media, particularly as a means of establishing and reinforcing relations between political leaders and their electorate, has meant a previously unimaginable amount of text data is now available for analysis. Moreover, the surge in the development of machine learning techniques – in particular, natural language processing – has given experts in the field the possibility to inspect this large amount of data with a precision and speed which were never possible before. One of the most influential among definitions of populism is the “ideational approach”, developed by scholars such as Mudde (2004) and Laclau (2005). A recent expert survey, conducted by Pippa Norris in collaboration with the Chapel Hill group (Norris, 2020), employed this approach to conduct the largest party classification to date concerning populism and the use of populist rhetoric. This thesis, using a dataset of 3.5+ million Tweets posted by politicians of 23 countries in the last 10 years, aims to explore the possibility of constructing a similar, if not equivalent, text-based definition of populism, employing machine learning and natural language processing techniques to extract features directly from text, rather than from the party affiliation of each leader, as in Norris' survey study.