Early identification of diffusion source in complex networks with evidence

The inferring of source in the aftermath of a pandemic outbreak has received immense attention due to its substantial practical potential. In light of the inaccessibility to the status of every individual in the network, observer-based approaches have become an essential research direction for solving this problem. However, the way that utilizes the combined observers to infer the source node, as most existing methods follow, may compromise the algorithm's flexibility and generalizability. To address this issue, we ask the question: can observational information be viewed as pairs of expert-driven information so that it allows us to frame the source localization as an information-fusion problem? To this end, we propose an evidential source localization (ESL) model that utilizes evidence theory to represent the uncertainty caused by limited information and to determine the source node by the information-fusion technique. Moreover, ESL is characterized by its ability to detect sources of disease at an early stage of the pandemic. Rather than focusing on a specific structure, we consider an arbitrary graph as our subject, rendering our method general enough for more applications. Experimental results on real-world networks demonstrate the superiority of ESL in efficiency and effectiveness with the comparison to other state-of-the-art methods.