{"ID":2836680,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.19837","arxiv_id":"2511.19837","title":"GED-Consistent Disentanglement of Aligned and Unaligned Substructures for Graph Similarity Learning","abstract":"Graph Similarity Computation (GSC) is a fundamental graph related task where Graph Edit Distance (GED) serves as a prevalent metric. GED is determined by an optimal alignment between a pair of graphs that partitions each into aligned (zero-cost) and unaligned (cost-incurring) substructures. Due to NP-hard nature of exact GED computation, GED approximations based on Graph Neural Network(GNN) have emerged. Existing GNN-based GED approaches typically learn node embeddings for each graph and then aggregate pairwise node similarities to estimate the final similarity. Despite their effectiveness, we identify a mismatch between this prevalent node-centric matching paradigm and the core principles of GED. This discrepancy leads to two critical limitations: (1) a failure to capture the global structural correspondence for optimal alignment, and (2) a misattribution of edit costs driven by spurious node level signals. To address these limitations, we propose GCGSim, a GED-consistent graph similarity learning framework centering on graph-level matching and substructure-level edit costs. Specifically, we make three core technical contributions. Extensive experiments on four benchmark datasets show that GCGSim achieves state-of-the-art performance. Our comprehensive analyses further validate that the framework effectively learns disentangled and semantically meaningful substructure representations.","short_abstract":"Graph Similarity Computation (GSC) is a fundamental graph related task where Graph Edit Distance (GED) serves as a prevalent metric. GED is determined by an optimal alignment between a pair of graphs that partitions each into aligned (zero-cost) and unaligned (cost-incurring) substructures. Due to NP-hard nature of exa...","url_abs":"https://arxiv.org/abs/2511.19837","url_pdf":"https://arxiv.org/pdf/2511.19837v1","authors":"[\"Zhentao Zhan\",\"Xiaoliang Xu\",\"Jingjing Wang\",\"Junmei Wang\"]","published":"2025-11-25T02:07:30Z","proceeding":"cs.LG","tasks":"[\"cs.LG\",\"cs.AI\",\"cs.DB\"]","methods":"[\"Graph Neural Network\"]","has_code":false}