Answers are grounded in this paper and cite their sources.
Entity alignment is a pivotal technology for integrating heterogeneous knowledge graphs, yet most embedding-based alignment methods destroy the inherent structural embedding information of knowledge graphs when unifying cross-graph embedding spaces, resulting in limited alignment accuracy and stability. To tackle this issue, this paper proposes a novel model for entity alignment tasks, which named as Graph Convolutional Network with Maximum Translated Similarity (GCN-MTS). GCN-MTS innovatively leverages a maximum translated similarity mechanism to optimize the adjacency matrix, degree matrix and loss function. This core design effectively maintains the original relative embedding structure of cross-graph entities during embedding space unification, which is the key improvement over existing approaches. Comprehensive experiments on 3 pairs of real-world knowledge graph dataset demonstrate the superior performance of GCN-MTS against advanced baseline methods. It achieves average performance gains of 5%–15% in Hits@1 and 4%–16% in Hits@10. Ablation results confirm that the optimized module brings a nearly 22% improvement in alignment metrics compared with the vanilla GCN. Furthermore, GCN-MTS presents excellent robustness and stability across different seed alignment quantities and seed combinations.
No takes yet. Share an insight, caveat, or question.
Jianhui et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: