Outcomes of the Award: 

This NSF funded RAPID project aimed to understand the psychological impacts of the COVID-19 pandemic, gleaning the information and misinformation that were and continue to be rampantly shared on social media and other online platforms. 

The project led to many successes. Included is a paper published in Nature Scientific Reports, which conducted a large-scale social media data driven study to assess how sharing COVID-19 related misinformation exacerbates anxiety, establishing an adverse psychological impact of misinformation, with a disproportionate impact on marginalized populations. A second paper from the team appeared at the premier web and social media conference, ICWSM 2022. This work uncovered that the adoption of large language models for building approaches for tasks aimed at detecting  crisis-related  information online, such as that around the pandemic, leads to systematically lower performance on non-English languages when compared to the performance on English. This indicates more research and infrastructural investments that will be needed to ensure that online discourse, spanning different languages, different populations, and different parts of the world, can be better monitored for rehabilitation purposes. A final paper is under preparation to be submitted to the journal, Nature Humanities and Social Science Communications. This is a systematic literature review that was spurred when we began research on this NSF project. We found that there was no comprehensive review of research that sought to understand the mental health impacts of crisis events. Since this was the sole goal of our project, we thus embarked on a path to fill this gap, to provide previously unknown insights into the relationship between protracted crises like COVID-19 and its longer-terms impacts on the well-being of populations directly or indirectly affected by the pandemic. 

Project Outcomes and Findings: 

Intellectual Merit: The project analyzed millions of social media postings in relation to the COVID-19 pandemic. In particular, we conducted an observational study involving novel causal analyses (such as propensity score matching) on a massive social media dataset comprising over 80 million Twitter posts by 76,985 users during an 18.5-month period. This established new empirical insights on the extent to which online sharing of COVID-19 misinformation results in worsened mental health outcomes. 

The project also resulted in important methodological takeaways within the natural language processing (NLP) and crisis informatics literatures. We empirically demonstrated that using images as an additional modality leads to a lesser difference between the performance on English and non-English text when it comes to machine learning detection of crisis related postings on social media.  

Finally, our research provided important domain-related intellectual takeaways for future researchers. Specifically, our systematic literature review of over 150 research articles in the crisis informatics space focused on unravelling the role that online platforms play by allowing users to share mental health concerns, and sentiments, building community resilience and finding supportive communities. 

Broader Impact: 

This project bears important implications toward the management and rehabilitation of global crisis events like the COVID-19 pandemic that are arguably increasingly frequent in the 21st century. First, in today's digital age, online misinformation accompanies almost every major crisis event. In the context of the COVID-19 pandemic, experts have speculated that exposure and consumption of misinformation can potentially worsen the mental health of individuals. Our findings showing that sharing misinformation exacerbates anxiety, establishing an adverse psychological impact of misinformation can inform social media platforms, policy-makers, and mental health caregivers, and can have a profound and lasting impact on the lives of many. Second, large language models are now the standard to develop state-of-the-art solutions for text detection and classification tasks. However, the development of advanced computational techniques is disproportionately focused on the English language. Our empirical findings that focus on 5 application domains with connection to crises highlight how existing language disparities could have possibly threatening implications on the lives of many individuals who belong to underserved communities. Third, our systematic and thorough review of nearly a decade of work in the field of crisis informatics identified how online data from users during crises has been harnessed. While our review aimed to provide a comprehensive guide for researchers and practitioners who intend to understand the impact of psychological well-being during crises via data from online platforms, we also noted several gaps in the existing research.

Aside from these contributions, the project also resulted in developing and furthering meaningful interdisciplinary collaborations that have resulted in successful subsequent fundraising. Findings of the research have also received coverage in popular news media such as the New York Times, and stemmed from the training of graduate students in Computer Science, Communication, and Psychology across three institutions, on an interdisciplinary topic of broad societal interest. Finally, we have built and released multiple open source tools and resources, ranging from datasets, human annotations, and machine learning classifiers, which we hope will support further investigation in this space.

Last Modified: 07/26/2022
Modified by: Munmun De Choudhury