Unmasking the Twitter Discourses on Masks During the COVID-19 Pandemic: User Cluster-Based BERT Topic Modeling...

Area of Science:

• Computational Social Science
• Public Health Informatics
• Political Communication

Background:

• The COVID-19 pandemic highlighted the politicization of public health, necessitating tools to analyze political contexts in public health discourse.
• Current infoveillance methods often overlook user identity and interests, limiting the understanding of how different political groups engage with public health topics.
• Algorithmic classification of users and their social media content can address these limitations by revealing nuanced discourse variations.

Purpose of the Study:

• To implement a novel computational framework for investigating discourse and temporal topic changes across distinct user clusters.
• To contextualize web-based public health discourse by analyzing variations across identity and interest-based user groups.
• To apply the framework to the case study of mask-wearing discourse during the early COVID-19 pandemic.

Main Methods:

• Clustering of Twitter users based on identity and interests derived from Twitter bio pages.
• Exploratory text network analysis to identify salient political, social, and professional identities within user clusters.
• BERT Topic modeling to identify temporal shifts and variations in discourse topics across four user clusters: conservative, progressive, general public, and public health professionals.

Main Results:

• User classification and longitudinal topic analysis are crucial for understanding the political context of public health discourse.
• Political groups and the general public focused on both the science and partisan politics of mask-wearing policies.
• Populist discourse, political figures (e.g., Donald Trump), and geopolitical tensions influenced public health discussions, with limited engagement from public health professionals.

Conclusions:

• A priori user classification is essential for effective analysis of web-based public health discourse.
• BERT Topic modeling is well-suited for identifying contextualized topics within short social media texts.
• Understanding user-specific discourse is vital for developing targeted and effective public health interventions in a politicized environment.