The CODE project focuses on studying the relationship between opinion dynamics, influenced by the dissemination of information and online debate, and the management of epidemics. On the one hand, it aims to shed light on the role of social media in the success of containment policies; on the other, the goal is understanding whether extreme and unexpected events, which are followed by government actions dictated by emergency, can lead to polarization of public debate and political radicalization. Inevitably, such research activity also requires investigating the nature of online and physical interactions, and the impact of any similarities and divergences between these two systems.
Discursive communities
We are studying the problem of detecting clear and meaningful communities in an online social debate. In particular, we are analyzing a framework consisting of 4 main steps: (i) distinguishing between main authors and audience; (ii) building a suitable informative network of authors; (iii) finding communities in this sub-network; (iv) propagating the labelling to the audience. This method is more effective than off-the-shelf community detection algorithms in retrieving ideological/political communities, and will be used to analyze how political affiliations influence information consumption and opinion dynamics on social media.
Meso-scale mixing patterns in geometric models
To define a suitable probabilistic network model that describes the physical interactions that enable disease transmission and, possibly, word of mouth, we are working on a better understanding of how meso-scale patterns can be enforced in geometric random networks. These models, in fact, have been shown to provide an elegant framework to generate networks with many desirable properties –e.g. small-worldness, high clustering and scale-free degree distribution– based on the combined effect of popularity and similarity. Our research work will show that, when a specific community structure is desirable, this must be enforced separately from the latent geometry of the network, and that it can be done with an entropy-based approach that guarantees that no other features of the network are inadvertently imposed.
Heterogeneous SIR — Main models
We are working on the formalization and analysis of a modified SIR model to study how heterogeneous behavior impacts on epidemic diffusion patterns. One key element of this model is that “ordinary” and “misbehaving” individuals differ by both their susceptibility and infectivity. Preliminary results show that the presence of a group of misbehaving individuals gives rise to a “resurgence zone”, where an initial fall in the size of the infected population is followed by a sudden growth that leads to an outbreak. A research paper describing the base model and the first set of analytical and experimental results will be presented to Complex Networks 2024. We are currently working on a dynamic extension of the model, where misbehaving individuals can turn ordinary if sufficiently many of their neighbors get the infection, to represent the effect of awareness on disease risks. The analysis, both theoretical and simulation-based, highlights a few non-trivial effects, such as a phase transition in the infection probability of individuals who made the transition.
Heterogeneous SIR — Data-driven model
We are also conducting a study based on the idea that misbehaving individuals are not uniformly spread in the population and localized clusters can impact the spread of a disease on a real territory. Using real data to define a synthetic population, whose individuals are characterized by a vector of features including geographic, socio-demographic and political attributes, we make the misbehaving probability depend on such features and use simulations on several Italian cities to investigate how the presence of inhomogeneity in the distribution of some attribute within and across cities is reflected in measurable differences in the spread of the epidemic.
Vaccination and waning immunity
While the latter is more focused on risk-perception and safe/unsafe interactions, in another ongoing work we are completing the analysis of a modified SIS model for diseases focused on vaccination, where coming into contact with the virus (whether through infection or vaccination) only provides temporary and waning immunity. Based on a simple model with just 3 levels of susceptivity and 2 levels of infectivity, we show that a non-trivial optimal vaccination rollout exists, provided that the vaccination capacity exceeds a precise threshold –under which only “first doses” should be distributed.