Vital nodes identification in temporal networks

Abstract

Temporal networks offer a powerful way to represent the dynamic behavior of many real-world systems, ranging from social interactions to communication infrastructures. Identifying vital nodes in these networks is important for a variety of applications, such as controlling epidemics, targeting marketing campaigns, or preventing cascading failures in power grids. Although many methods have been developed to find influential nodes in static networks, extending them to temporal networks remains challenging. This difficulty becomes even more pronounced when privacy restrictions limit the amount of available data. This thesis explores Graph Neural Network (GNN)-based approaches for vital node identification (VNI) in temporal networks. I propose a method that learns temporal node embeddings from a sequence of networks and then aggregates these embeddings using an attention mechanism over time steps. This design allows the model to highlight the most relevant moments for node influence, rather than treating all time steps equally. Finally, the model predicts a vitality score for each node based on its aggregated embedding. Experiments on several real-world datasets show that the proposed method can outperform a baseline approach in specific settings, highlighting the potential of GNNs for temporal node analysis. However, the results also reveal some important limitations. Indeed, the proposed approach depends on Suspected-Infected-Recovered (SIR) based simulations to generate ground-truth vitality scores, which are computationally demanding and sensitive to parameter choices. Additionally, the model’s performance strongly relies on hyperparameter tuning and the characteristics of the dataset used for training and evaluation.