Network Alignment Using Graphlet Signature and High Order Proximity

Network alignment problem arises in graph-based problem formulation of many computer science and biological problems. The alignment task is to identify the best one-to-one matching between vertices for a pair of networks by considering the local topology or vertex attributes or both. Existing algorithms for network alignment uses a diverse set of methodologies, such as, Eigen-decomposition of a similarity matrix, solving a quadratic assignment problem through subgradiant optimization, or heuristic-based iterative greedy matching. However, these methods are either too slow, or they have poor matching performance. Some existing methods also require extensive external node attributes as prior information for the purpose of node matching. In this paper, we develop a novel topology-based network alignment approach which we call GraphletAlign. The proposed method uses graphlet signature as node attributes and then uses a bi-partite matching algorithm for obtaining an initial alignment, which is later refined by considering higher-order matching. Our results on large real-life networks show the superiority of GraphletAlign over the existing methods; specifically, GraphletAlign’s accuracy improvement is up to 20%–72% compared to existing network alignment methods over six large real-life networks.