Which Network Similarity Measure Should You Choose: An Empirical Study

We consider the problem of determining how similar two networks, without known node-correspondences, are. This problem occurs frequently in real-world applications like transfer learning and change detection. Many network similarity measures exist, and it is unclear how one might select from amongst them. We provide the first empirical study on the relationships between different network similarity methods. Here, we propose (1) an approach for identifying groups of comparable network similarity methods and (2) an approach for computing the consensus among a given set of network similarity methods. We apply our approaches to seven real datasets and twenty network similarity methods. Our experiments demonstrate that (a) different network similarity methods are surprisingly well correlated, (b) some complex network similarity methods can be very closely approximated by much simpler methods, and (c) two network similarity methods–namely, random walk with restarts and NetSimile–provide similarity rankings that are closest to the consensus ranking.