Detecting Multiple Epidemic Sources in Network Epidemics Using Graph Neural Networks

Rodrigo Gonçalves Haddad, Daniel Ratton Figueiredo

Abstract: Epidemics start within a network because of the existence of epidemic sources that spread information over time to other nodes. Data about the exact contagion pattern among nodes is often not available, besides a simple snapshot characterizing nodes as infected, or not. Thus, a fundamental problem in network epidemic is identifying the set of source nodes after the epidemic has reached a significant fraction of the network. This work tackles the multiple source detection problem by using graph neural network model to classify nodes as being the source of the epidemic. The input to the model (node attributes) are novel epidemic information in the k-hop neighborhoods of the nodes. The proposed framework is trained and evaluated under different network models and real networks and different scenarios, and results indicate different trade-offs. In a direct comparison with prior works, the proposed framework outperformed them in all scenarios available for comparison.