Topic Modelling and Frame Identification for Political Arguments

SHOHREH HADDADAN; ELENA CABRIO; AXEL J. SOTO; VILLATA, SERENA

Udine

Conferencia; 21st International Conference of the Italian Association for Artificial Intelligence; 2022

Associazione Italiana per l?Intelligenza Artificiale

Presidential debates are one of the most salient moments of a presidential campaign, where candidates are challenged to discuss the main contemporary and historical issues in a country. These debates represent a natural ground for argumentative analysis, which has been always employed to investigate political discourse structure in philosophy and linguistics. In this paper, we take the challenge to analyse these debates from the topic modeling and framing perspective, to enrich the investigation of these data. Our contribution is threefold: first, we apply transformer-based language models (i.e., BERT and RoBERTa) to the classification of generic frames showing that these models improve the results presented in the literature for frame identification; second, we investigate the task of topic modelling in political arguments from the U.S. presidential campaign debates, applying an unsupervised machine learning approach; and finally, we discuss various visualisations of the identified topics and frames from these U.S. presidential election debates to allow a further interpretation of such data.