Install the app

Install this application on your home screen for quick and easy access when you’re on the go.

Just tap Share then “Add to Home Screen”

Political Research Exchange - PRX

Profiling Malignant Rhetoric: Linking Cognitive Linguistics and Machine Learning Algorithms to Evaluate the Emotionality of Populist Discourse within Social Networks

Social Media
Big Data
Annika Hamachers
German University of the Police
Annika Hamachers
German University of the Police
Andrej Galic
German University of the Police

Since emotions are considered to be an important precondition for persuasion, it is almost common place knowledge nowadays, that populists deliberately employ rhetorical strategies (e.g. metaphors, figurative language, see Boeynaems, Burgers, & Konijn, 2018) that aim at an emotionalization of public topics.
Schneider et al. (2018) for instance find fear of crime to be among the predominant themes in Islamist and right-wing YouTube videos. Other research even suggests that a rhetoric which combines a specific set of emotions (namely anger, contempt, and disgust, as proposed by the ACONDI model) has the power to mobilize people and turn feelings of dissatisfaction into violent action (Matsumoto et al. 2012; 2015) or so called ‘vigilante terrorism’ (Quent, 2016; 2017).
However, it remains unclear how exactly, to what extent, and in which combination emotions are elicited in populist contents.
In our current work we try to link dictionary-based approaches from cognitive linguistics with machine learning techniques to assess the emotionality of populist expressions on social media:
We heavily rely on the work of James Pennebaker who compiled and validated a large set of word lists (so-called ‘dictionaries’) representing psychometric features of the authors that make use of these words (including categories for the expression of anger and anxiety) (Pennebaker, 2015; Boyd & Pennebaker, 2017). Additionally, we borrowed lists corresponding to disgust, joy, and trust from the NRC (National Research Council Canada) emotion dictionary (Mohammad & Turney, 2013). Finally, we enhanced these pre-existing collections of words by performing supervised and unsupervised topic modelling on our data corpus – creating an algorithm that looks out for emotion expressions that are ‘domain specific’ for populist discourse. These final lists at hand, we are able to profile the emotional nature of populist discourse, trace the relation of different emotions across time, and gather support for the ACONDI hypothesis.
Share this page