NF-Align: A Neighborhood-Focused Alignment in Large-scale Communication Networks

Alignment of entities in large-scale communication networks aims to match equivalent entities across multiple networks. Most existing approaches aggregate information of direct neighbors using graph neural networks. Nevertheless, considering all direct neighbors may introduce noise, as some neighbors are less informative for alignment. We propose a Neighborhood Focused Alignment (NF-Align) model to enhance a node’s important neighbors by removing its less useful neighbors. First, we assign importance values to a node’s neighbors based on a combination of node degree and motif-based node degree, which measures a neighbor’s participation in communication networks. Then, we construct a neighborhood-focused graph by removing edges of less important neighbors. The focused graph retains fewer but more engaged neighbors. When used as a plug-in module, the neighborhood-focused graph can refine the neighborhood before feeding into any alignment model. Experiments on benchmark datasets demonstrate that NF-Align improves the performance of original models and outperforms state-of-the-art baselines by removing less important edges and reducing noise. Our model provides a novel approach to aligning entities in large-scale communication networks by emphasizing more engaging neighbors.