Structured Interactions - Inferring Social Behaviour in Networked Systems

Social behaviour permeates our lives. We live locally through our friends, family and acquaintances, and on larger scales by sharing cultural values, identities and political opinions. Sociology has a rich tradition of characterising societal dynamics through concepts such as social roles, group identities and social spaces. At the same time, network science offers a valuable framework for analyzing interactions patterns, integrating tools from physics and data analysis to connect micro-level contacts with large-scale phenomena. Aided by the digital traces left behind by information technologies, both fields have helped us understand how we communicate, how information and epidemics spread in social systems and provided insights into polarization. These advances have revealed that the structures and dynamics of social systems are deeply intertwined, prompting broader questions about how social behaviour affects and is affected by networked phenomena across scales. Our contributions elucidate how networked social behaviour is reflected on empirical communication data and how it can be modeled through network dynamics —two pillars of modern network science. We evaluate communication patterns through the lenses of longstanding sociological theories, revealing how tie strength and multiplexity —overlapping social roles— manifest in rich temporal data. We analyse features of temporal communication that act as indicators of relationship strength, assessing their prevalence and limitations across various communication mediums. We also show that multiplexity has a temporal expression through the use of different social times during the week, such as worktimes and weekends. We show that such multiplexity has structural effects on networks that align with the expectations of the theory of social foci. We also assess the degree to which these communication patterns mirror intrinsic human behaviour by analyzing a historical dataset of epistolary communication. Then, we model how mechanistic networked processes such as meeting friends-of-friends or popular people can interact with group identities and lead to salient large-scale phenomena, such as polarization or inequality between groups. We analyse models of choice homophily, triadic closure and preferential attachment, showing that the latter two can impose structural constraints —inducing more in-group ties and leading to core-periphery networks where one group dominates connections, respectively. We use different inferential frameworks to quantify induced homophily in empirical systems, and to infer the presence of these mechanisms in data with core-periphery parings. Our findings reveal how forms of social behaviour function as both drivers and outcomes within networked systems. The modelling and inferential methodologies we propose uncover a wide spectrum of societal patterns, encompassing tie strength, multiplexity, and structural biases induced by homophily and networked interactions.