Investigating Canadian Public Attitudes Toward COVID-19 Vaccine Mandates with a Nested Analysis Framework

Area of Science:

• Computational Social Science
• Natural Language Processing
• Public Health Informatics

Background:

• Social media, specifically Reddit, is crucial for understanding public sentiment towards COVID-19 vaccine mandates in Canada.
• Network communities on Reddit offer valuable data for analyzing public attitudes.

Purpose of the Study:

• To develop and evaluate a computational framework for analyzing public discourse on COVID-19 vaccine mandates using Reddit data.
• To identify key themes and topics within online discussions related to vaccine mandates.

Main Methods:

• A nested analysis framework was employed, collecting 20,378 Reddit comments via the Pushshift API.
• A BERT-based binary classification model was developed to screen for comment relevance, achieving 91% accuracy.
• A Guided Latent Dirichlet Allocation (LDA) model was applied to relevant comments for topic extraction, identifying four key topics with 83% human-evaluated accuracy.

Main Results:

• Out of 20,378 comments, 3,179 (15.6%) were relevant to COVID-19 vaccine mandates.
• The BERT model demonstrated high accuracy in filtering relevant comments.
• The Guided LDA model successfully extracted four main topics: travel, government, certification, and institutions.

Conclusions:

• A novel screening and topic modeling tool was developed for analyzing Reddit comments on COVID-19 vaccine mandates.
• The study highlights the utility of NLP techniques in understanding public opinion on health policies.
• Future work should focus on refining seed word selection and evaluation methods for topic modeling to minimize human oversight.