Similarity Evaluation on Labeled Graphs via Hierarchical Core Decomposition

Abstract

Graph similarity is essential in network analysis and has been applied to various fields. In this paper, we study the graph similarity between labeled graphs, i.e., every vertex is assigned to a label. Since few methods take account of the structure of a graph and most existing methods cannot extend to massive graphs, we develop a novel graph similarity measure that overcomes the above limitations. Given two labeled graphs, our proposed method first utilizes the concept of k-core to organize the connected cohesive subgraphs of each graph in a tree-like hierarchy. Then, the graph similarity between them is computed from their tree-hierarchies. An efficient algorithm is also developed for the proposed measure. Extensive experiments are conducted on 6 public datasets, where our proposed algorithm successfully identifies similar graphs and extends to large-scale graphs.