.png){kind=icon}
Return
Adaptive multi-embedding framework for unsupervised network alignment
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2026.116755.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Network alignment aims to identify node correspondences across different networks and plays an important role in many downstream tasks. Due to the difficulty of obtaining prior anchor links, unsupervised network alignment has attracted increasing attention. However, many existing unsupervised methods have limited ability to capture complex structural patterns or rely on fixed fusion ratios to combine multiple sources of information, which restricts their performance and makes them vulnerable to noise. To address these limitations, we propose an unsupervised alignment framework called AMENA, which comprehensively models node structural information from local connections to high-order roles and adaptively integrates multiple node representations. Specifically, we first model node information from three complementary perspectives: attribute features, local structural features, and high-order role features. These features are fed into shared-weight GATv2 models to generate corresponding node embeddings. Additionally, we design an adaptive aggregation mechanism that estimates the reliability of each embedding according to the number of trusted pseudo-anchor pairs it produces, enabling the model to emphasize more reliable information sources across networks. Finally, an iterative refinement process is applied to further enhance alignment accuracy by improving matched neighborhood consistency. Extensive experiments on real-world and synthetic datasets demonstrate that AMENA outperforms state-of-the-art baselines and exhibits strong robustness under various challenging conditions.
Keywords:
Unsupervised network alignment
GAT
Multi-embedding
Adaptive aggregation
Graphlets
Journal
K
IF:
7.6
Papers:
1.3W
Citations:
4.5W
Organization
Cited Papers
No cited papers available