Human social interactions take place across diverse settings—ranging from face-to-face encounters in physical spaces to exchanges mediated by digital platforms. These different environments generate data with distinct structural and temporal properties, demanding tailored modeling approaches. “This thesis explores how to represent, analyze, and simulate human interaction dynamics across physical and digital environments. In the physical realm, we introduce a novel higher-order network framework that captures group interactions and temporal structure beyond pairwise links. In the digital context, we present a case study on decentralized online social networks, focusing on how moderation dynamics shape long-term structure and information flow.” In the first part of the thesis, we focus on physical social systems and propose a framework for characterizing local interaction patterns as temporal neighborhoods enriched with higher-order (e.g., triadic) interactions. This approach enables both the comparison of individuals and the classification of behavioral dynamics across different social settings.
We then extend this framework to develop a generative model capable of producing surrogate temporal networks that reproduce the group structure and time-varying behavior observed in empirical data. The model demonstrates improved fidelity over first-order baselines, especially in capturing clustering and higher-order connectivity patterns. In the second part, we turn to online decentralized platforms, using Mastodon as a case study to understand how moderation and governance shape network evolution and information spread. By modeling inter-instance interactions as a signed and temporal network, we show that despite active moderation and structural change, the system exhibits long-term stability and equilibrium-like properties. Simulations of information diffusion reveal that moderation enforces asymmetric communication pathways and induces echo chamber effects—even in the absence of centralized control. Together, these results offer a unified perspective on human interaction dynamics across both physical and digital domains, emphasizing the importance of temporal structure and higher-order modeling in understanding complex social systems.