Moralized Social Media in the 2020 Presidential Election

• In Event: Poster Session: Political Methodology & Political Science Education & Civic Engagement
  In Poster Session: Poster Session: Political Methodology

Abstract

This study examines the moralization of political discourse on social media over the 2020 US presidential election campaign trail and explores the relationship between moralization and partisanship. Commentators have observed the rise of moralistic rhetoric in contemporary American politics. Concrete issues in public policy are increasingly interpreted with categorical judgements of right and wrong. A moralized political climate leaves little room for deliberation on the multifacetedness of concrete policy issues and solidifies uncompromising attitudes. Taking a closer look at the moral discourse in the context of a presidential election campaign trail will shed light on the process of moralization, whether it reinforces or descends from political polarization.
This study applies the state-of-art transformer-based models to detect moral content in social media posts. A BERT-based classifier was trained on an annotated moral foundations Twitter corpus (Hoover et al., 2020). BERT is a pretrained deep learning language model. The classifier was fine-tuned on a pretrained BERT model, after comparing the performances of different versions of pretrained models. It achieves 84% precision in moral content detection and outperforms the relatively traditional machine learning models such as SVM regression. It achieves a decent rate of overall accuracy (nearly 80%) for such a relatively difficult task.
The classifier was then used for inferring the moral content on a large-scale Twitter corpus on the 2020 presidential election that is publicly available (Chen et al., 2020). Classification of moral content was conducted on Tweets from the start of March to the end of November in 2020. Analyses of outstanding moralized topics leading up to the 2020 election are conducted including mail-in voting, covid-19 and voter fraud. Preliminary analyses indicate that moralization process happens fast on social media and that moralized discourse is reinforced through periodic peaks. How these topic frames evolved and how users of different partisan camps helped shape the narrative evolution will be further examined through semantic network analysis.

Author

• Meiqing Zhang, University of Southern California