In the global pandemic era of COVID-19, a time filled with high risk and uncertainty, people become more vulnerable than ever to be influenced by problematic information, such as messages provoking strong emotions, discrimination, violence, and distrust. How do such problematic signals accumulate and translate into a toxic influence over online space?  This project aims to study the online ecosystem that fertilizes the "infodemic" in the COVID-19 outbreak, focusing on the characteristics of misinformation content and consumers. By leveraging computational social science methodology and multi-modal learning, this interdisciplinary research makes several contributions to developing methods to understand the proliferation and persuasive narrative construction of misinformation.

First, ​​we developed a computational approach to analyze the patterns of persuasive social media content, in terms of popularity and source credibility. Our multi-modal approach, combining image and text, is not only predictive of information popularity and credibility, but also able to uncover how unreliable sources integrate visual elements with textual content in a distorted, biased fashion. This result provides insights into the enhancement of social media literacy and engagement.

Second,  through analyzing a large, geopolitically diverse sample of Twitter users and their information consumption over a 6-month period, we identified "who" constitutes the population vulnerable to the online misinformation in the pandemic, and what are the robust features and short-term behavior signals that distinguish susceptible users from others. We discovered that (1) contrary to the prior studies on bot influence, our analysis shows that social bots' contribution to misinformation sharing was surprisingly low, and human-like users' misinformation behaviors exhibit heterogeneity; (2) susceptible users appeared to be politically sensitive, active, and responsive to emotionally charged content among susceptible users. In terms of predicting users' susceptibility, we developed an interpretable deep learning model that efficiently forecasts users' transient susceptibility solely based on their short-term news consumption and exposure from their networks. Our results contribute to designing effective intervention mechanisms to mitigate the misinformation dissipation.

Third, through an analysis of over 240 thousand tweets capturing how users shared COVID-19 pandemic-related misinformation news on social media over a five-month period, we study the spread of information and user's level of interaction with the original information source, i.e., from low (copy-and-paste link) to medium (Like), to high (comments). The higher the user's interacting with the original source, the more mutation the spread of information is. Our results indicate a positive relationship between information mutation and spreading outcomes. This study provides the first quantitative evidence of how misinformation propagation may be exacerbated by users' commentary, which has implications for countering misinformation.

Finally, this project has two broader impacts. First, our study yields promising results by characterizing the information ecosystem, which enriches the social solutions in countering misinformation proliferation beyond the fact-checking paradigm. Second, the longitudinal data and techniques we developed enable researchers, platform managers, and policy analysts to discover new insights into information consumption behaviors and test new methods to counter misinformation.

Last Modified: 06/30/2022
Modified by: Yu-Ru Lin