1. Network Infusion to Infer Information Sources in Networks
- Author
-
Gerald Quon, Ken R. Duffy, Manolis Kellis, Soheil Feizi, and Muriel Medard
- Subjects
FOS: Computer and information sciences ,Physics - Physics and Society ,Computer Networks and Communications ,Computer science ,FOS: Physical sciences ,Inference ,Physics and Society (physics.soc-ph) ,02 engineering and technology ,computer.software_genre ,0202 electrical engineering, electronic engineering, information engineering ,Diffusion (business) ,Social and Information Networks (cs.SI) ,Social network ,business.industry ,Computer Science - Social and Information Networks ,020206 networking & telecommunications ,Inversion (meteorology) ,Inverse problem ,Computer Science Applications ,Control and Systems Engineering ,Homogeneous ,020201 artificial intelligence & image processing ,Minification ,Data mining ,business ,Centrality ,computer - Abstract
Several significant models have been developed that enable the study of diffusion of signals across biological, social and engineered networks. Within these established frameworks, the inverse problem of identifying the source of the propagated signal is challenging, owing to the numerous alternative possibilities for signal progression through the network. In real world networks, the challenge of determining sources is compounded as the true propagation dynamics are typically unknown, and when they have been directly measured, they rarely conform to the assumptions of any of the well-studied models. In this paper we introduce a method called Network Infusion (NI) that has been designed to circumvent these issues, making source inference practical for large, complex real world networks. The key idea is that to infer the source node in the network, full characterization of diffusion dynamics, in many cases, may not be necessary. This objective is achieved by creating a diffusion kernel that well-approximates standard diffusion models, but lends itself to inversion, by design, via likelihood maximization or error minimization. We apply NI for both single-source and multi-source diffusion, for both single-snapshot and multi-snapshot observations, and for both homogeneous and heterogeneous diffusion setups. We prove the mean-field optimality of NI for different scenarios, and demonstrate its effectiveness over several synthetic networks. Moreover, we apply NI to a real-data application, identifying news sources in the Digg social network, and demonstrate the effectiveness of NI compared to existing methods. Finally, we propose an integrative source inference framework that combines NI with a distance centrality-based method, which leads to a robust performance in cases where the underlying dynamics are unknown., Comment: 21 pages, 13 figures
- Published
- 2019