Leveraging Structural Hierarchy for Scalable Network Comparison

K-core decomposition is a popular method that segments a network revealing the underlying hierarchy. We explore the propensity of this decomposition method for structural discrimination among networks by extracting features from each level of the hierarchy. We propose a novel algorithm for Network Comparison using k-coreDecomposition NCKD. The method is effective, efficient and scalable, with computational complexity of $$O|{\mathcal E}|$$, where $${\mathcal E}$$ is the set of edges in the network. The low computational complexity of the method makes it attractive for scalable network comparison. NCKD algorithm decomposes networks and extracts features from the resulting shells. Jensen-Shannon distance between extracted features quantifies structural differences between networks. We establish that probability distributions of coreness and intra/inter-shell edges are capable of characterizing different genres of networks and capturing finer structural differences between networks of the same genre. We experiment with synthetic and real-life networks upi¾?to eight million edges on a single PC. Comparison with two recent state-of-the-art network comparison methods affirms that NCKD outperforms in terms of effectiveness and scalability.