Source identification of network diffusion processes with partial observations

Abstract

In this paper, we study the problem of identifying the source of network diffusion processes by using the contact networks and sampled infections. The dynamic processes are described by the susceptible-infected (SI) model and a maximum likelihood (ML) estimator is formulated for identification of the source. We propose a two-stage method that first locates a set of suspected source nodes and then identifies the infection source from the candidate source nodes by using Markov random field method. Experiments on regular tree and real-world networks verify the effectiveness of our approach.