Classifying Quality Centrality for Source Localization in Social Networks

Abstract

Source localization, the process of estimating the originator of an epidemic outbreak or rumor propagation in a network, is an important issue in epidemiology and sociology. With the graph topology of the underlying social network, the localization can be realized with observations of a few designated nodes or a snapshot of the whole network at a certain time. Though there are several methods for this task, all of them have limitations. These approaches either place little weight on information about susceptible nodes or rely on extra information about the propagation process. In this paper, we take both susceptible and infected nodes into account, and put forward a novel metric called Classifying Quality (CQ) centrality to quantify the property of a node to separate the susceptible and infected sets. Inspired by Fisher criterion, CQ centrality makes a trade-off between the inner-class and the inter-class distances, which are based on length of the shortest path between nodes. CQ centrality can be calculated without any extra information about the spread process, hence, it can serve as a universal estimator for source localization. Moreover, we improve the proposed metric in case that the infection rates of edges have been already known. Simulation results on various general synthetic networks and real-world networks indicate that our methods lead to significant improvement of performance compared to existing approaches.