Quantifying and Modeling risks and dynamics of information disorders in digital ecosystems

Social media platforms, as primary sources of information, have profoundly transformed public communication, reshaping how information is created, disseminated and consumed. While enabling unprecedented connectivity, they also amplify the risks associated with information disorders, ranging from the spread of misleading content to harmful online behaviors. The thesis examines the problem of information disorders by investigating four aspects: the characterization of online conversation during particularly important events, the measurements and validation of indexes able to quantify the risks of infodemic and the modeling of infodemic dynamics and radicalization in online social media. In particular, this dissertation provides a new framework capable of measuring and modeling how unreliable information spreads, how emotional and socio-economic factors influence such diffusion, and how these mechanisms affect the stability and resilience of digital ecosystems. To address this problem, we combined large scale data analysis, network science approaches and modeling to quantify and interpret the dynamics of information disorders. The first part focuses on the emotional responses associated with the online discussion of highly debated topics, such as the COVID-19 vaccination campaign and Long COVID. The second part proposes a new methodological approach to quantify the news media diet and information consumption patterns of users. Building upon existing indexes of infodemic risk, the third part provides a validation of such indexes by using socio-economic indicators, that explain cross country differences in the exposure and susceptibility to misinformation. In addition, a stochastic model is developed to reproduce the empirical volatility of the Infodemic Risk Index, revealing differences in the superspreader dynamics across countries. The thesis also extends its analytical framework to the study of radicalization dynamics on Reddit, showing how users’ linguistic and emotional patterns evolve after exposure to misogynistic communities, thus offering predictive insights into harmful online behaviors. This approach advances the state of the art by moving from descriptive analyses of misinformation toward a unified and quantitative modeling framework that connects micro-level user behavior with macro-level systemic instability. This thesis contributes to the development of the computational social science field, proposing a data-driven science of informational risk aimed at supporting policy design and digital governance for healthier and more resilient information ecosystems.