Models and data for behavioural spreading in social networks

Social networks have been widely studied both from the social sciences and from the mathematical point of view. Today, with the abundance of data at our disposal, it is possible to merge these fields of science to deepen our knowledge on social behaviours. In this thesis, I investigate the application of social-dynamics models to real-world scenarios. Social dynamics could be described by different theoretical models. A first class of models is the epidemiological one, which was first introduced to describe and predict the trends of epidemics caused by viral infections, but since then it has been extended also to other kind of social contagion. Then I discuss opinion dynamics models, used to describe political views and focus on the conditions that lead to consensus around certain opinions. Subsequently, I discussAgent-Based Models (ABMs), which can be used in a wide range of domains, and can incorporate features of both previous modelling approaches, offering greater flexibility at the cost of limited predictive power due to calibration challenges and their intrinsic stochastic nature. Following this theoretical foundation, I propose three applications across different contexts. The first model I develop concerns the diffusion of sentiment on online social platforms. In particular, I analyse Reddit communities focused on finance, reconstruct the underlying user networks, classify posts to infer user-level sentiment, and compare empirical patterns with an ABM that integrates epidemiological dynamics and trust between agents.I show that the proposed model successfully reproduces the sentiment trends observed in these communities The second application is related to a public health problem, in particular addressing tobacco consumption as a formof social contagion. Starting from this modelling, I apply an influence maximisation approach using the zealot node technique to improve the effectiveness of school-based interventions aimed at reducing smoking under budget constraints. Counter-intuitively, I find that targeting the low-degree nodes results in a greater reduction in the prevalence of smokers. Moreover, when applying this strategy, I find an optimal fraction of student to target, which is surprisingly lower than the full population. Finally, the last work again focuses on opinion formation, but emphasises the importance of network topology rather than dynamic mechanism. In this study, the network of living cardinals of the Roman Catholic Church is reconstructed and, after extracting and evaluating their opinions using large language models, the relationship between opinion similarity and network position is investigated, leading to identifying the relevance of specific relational structures on opinion formation. Overall, this thesis shows how combining theoretical models of social dynamics with data-driven analyses can yield valuable insights into collective behaviour across diverse contexts. By integrating epidemiological approaches, opinion dynamics, and agent-based modelling, thiswork highlights both the potential and the limitations of current modelling frameworks, while offering practical applications ranging from online sentiment diffusion to publichealth interventions and opinion formation in institutional networks