BENCHMARKING THE INFLUENTIAL NODES IN COMPLEX NETWORKS

Abstract

Among diverse topics in complex network analysis, the idea of extracting a small set of nodes which can maximally influence other nodes in the network has a variety of applications, especially for e-marketing and social networking. While there is an abundance of heuristics to identify such influential nodes, the method of quantifying the influence itself, has not been investigated in the research community. Most of the classical and state-of-the-art works use Diffusion tests for influence benchmark of a particular set of nodes in the network. The underlying study challenges this method and conducts thorough experiments to show that for real-world applications, the diffusion test alone is not only insufficient, but in some cases is also an inaccurate method of benchmarking. Using eight widely adopted heuristics, 25 networks were tested using Diffusion tests and compared with resilience test, we found out that no single algorithm performs consistently on both types of tests. Thus, we conclude that a more accurate way of benchmarking a set of influential nodes is to run diffusion tests alongside resilience test, in order to label a certain technique as best performer.