MAXIMIZING INFLUENCE OF SIMPLE AND COMPLEX CONTAGION ON REAL-WORLD NETWORKS

Abstract

Contagion spread over networks is used to model many important real-world processes from a wide variety of domains including epidemiology, marketing, and systems engineering. A large body of research provides strong theoretical guarantees on simple contagion models, but recent research identifies many real-world processes that feature complex contagions whose spread may depend on multiple exposures or other complex criteria.

We present a rigorous study of real-world and artificial networks across simple and complex contagion models. We identify domain-dependent features of real-world networks extracted from publicly-available networks as a guide to solving contagion-related decision problems. We then examine the performance of multiple influence-maximization algorithms across a space of networks and contagion models to develop an experimentally justified guide of best practices for related problems. In particular, genetic algorithms are an extremely viable candidate for these problems, especially with complex graphs and processes.