A Survey of GNN-Based Graph Similarity Learning

With the advancement of deep learning, the application of graphs has become increasingly widespread. Graph similarity learning has gained significant attention as a crucial component in various learning tasks, such as classification, clustering, and subgraph matching, making it a research hotspot in the field of graphs. Currently, graph similarity learning based on graph neural networks (GNNs) primarily involves mapping input graphs to a target space using deep learning models, aiming to approximate the structural distances between graphs in the output space. Firstly, this paper provided a comprehensive review of research on GNN-based graph similarity learning. Based on different graph representation learning approaches, it categorized them into three types: GNN-CNN hybrid models for graph similarity learning, Siamese GNN-based graph similarity learning, and GNN-based graph matching networks for similarity learning. Secondly this paper then provided a detailed analysis of these three types of models. Finally, it discusses the challenges and future research directions in GNN-based graph similarity learning.