Drivers of online abuse against Canadian public health officials: An LLM-based temporal analysis

This study identifies trends in abusive discourse towards public health professionals (PHPs) during the COVID-19 pandemic and explores associations between abusive digital content, case numbers, deaths, and major policy announcements.

Natural language processing (NLP) and large language model (LLM) techniques were used to develop a computational model to detect abusive content on X (formerly Twitter). This model was applied to abusive posts targeting PHPs during COVID-19 by examining over 1.7 million posts from January 2020 to May 2021. Associations between spikes in abuse and the number of COVID-19 cases, deaths, and government monitoring updates were explored.

Pronounced surges in abusive posts coincided with rising case and death counts and the imposition of major federal COVID-19 policies, particularly during the pandemic's initial emergency response. Digital aggression increased at times of public health statements related to restrictions in social gatherings, testing criteria, vaccinations, and masking. However, during high-case periods, even statements providing case updates, expressing compassion, or urging collective responsibility coincided with higher numbers of abusive posts.

This work provides insights into the pressures faced by health officials online and offers implications for designing resilient public communication during future crises.