Abstract :
[en] The widespread use of social media has transformed how information is created, shared, and consumed, but it has also enabled the rapid diffusion of misinformation with serious consequences for public health, democratic processes, and social trust. Despite extensive research on misinformation, important gaps remain in understanding how users engage with misleading content during health crises, how fact-checking organizations select claims for verification, and how effective emerging fact-checking interventions are in real-world settings. This thesis addresses these gaps by examining misinformation diffusion and fact-checking practices on social media, with a particular focus on the role and effectiveness of crowdsourced fact-checking approach, i.e., the Community Notes system deployed on X (formerly Twitter).
First, the thesis advances the understanding of user engagement with misinformation during health crises. Using the COVID-19 pandemic as a case study, it shows that misinformation often spans multiple topics and incorporates conspiracy narratives, which increases its attractiveness to users. Crucially, the thesis distinguishes between different user types on social media -- those who share external news items and those who react to them on X -- and demonstrates that these groups exhibit different engagement patterns. While news sharers contribute to the persistence of misinformation over time, post viewers are more likely to engage with content characterized by topic diversity and conspiratorial framing.
Second, the thesis examines traditional fact-checking practices by investigating how professional fact-checking organizations select politicized statements for verification. Analyzing fact-checked claims mentioning U.S. political elites over a long time span, it uncovers systematic differences between fact-checked true and false statements, partisan asymmetries, and temporal patterns around election periods. These findings shed light on how politicization and emotional framing could influence fact-checking priorities, contributing to debates about neutrality and trust in professional fact-checking.
Third, the thesis provides a comprehensive evaluation of community-based fact-checking through a series of large-scale empirical studies of X's Community Notes system. It offers causal evidence that community notes can substantially (61.2%) reduce the spread of misleading posts and increase the likelihood (94.3%) that authors delete their misleading posts once notes are displayed. However, it also shows that these effects are limited at the platform level because notes often appear too late to intervene during the early and most viral stages of diffusion. In addition, the thesis demonstrates that community fact-checks elicit strong emotional reactions, including increased negativity and moral outrage, highlighting both their corrective power and their role as signals of social norm violations.
Finally, the thesis investigates the algorithm resilience and participation expansion of Community Notes system, exploring how community-based fact-checking systems can be improved. It examines the stability of the consensus-based algorithm used to surface helpful notes and reveals vulnerabilities to post-display polarization and strategic rating behavior. It also evaluates the impact of expanding participation through user-initiated requests for fact-checks, identifying both the promise of scalable, user-led triage and the challenges arising from partisan misalignment between requestors and contributors.
Overall, this thesis makes important contributions to misinformation research by (i) advancing the understanding of misinformation diffusion, (ii) identifying partisan asymmetries in traditional fact-checking practices, (iii) evaluating the effectiveness of community-based fact-checking through a comprehensive and multi-dimensional perspective, and (iv) examining the algorithm resilience and participation expansion of community-based fact-checking. Together, these findings provide actionable insights for social media platforms to develop more reliable, timely, and resilient fact-checking systems.
