Network alignment
🤖 AI Summary
Network alignment aims to identify cross-network correspondences among entities in multiple complex systems, yet existing approaches suffer from domain isolation and terminological inconsistency. This paper proposes a differentiable topological alignment framework that— for the first time—integrates graph neural networks, optimal transport, and contrastive learning into a unified, end-to-end, unsupervised node mapping paradigm. The method jointly preserves local neighborhood consistency and global structural fidelity. It is applicable to both homogeneous and heterogeneous networks. Evaluated on multiple benchmark datasets, it achieves an average 7.2% improvement in alignment accuracy over state-of-the-art methods. Moreover, it exhibits strong generalizability and computational efficiency. By unifying modeling assumptions across domains, the framework establishes a scalable, principled foundation for cross-domain network analysis.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Identifying entity relationships across complex networks for alignment.
Comparing network alignment methods across diverse domains and conditions.
Addressing challenges in unifying network alignment research terminologies.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Structure consistency-based network alignment methods
Network embedding-based alignment techniques
GNN-based methods for network alignment
🔎 Similar Papers
No similar papers found.
