Abstract:Network alignment is a fundamental problem in various domains since it can establish bridges for the same entity (i.e., anchor nodes) between different networks. Most existing network alignment methods are based on consistency assumption, i.e., anchor nodes exhibit similar local structures or neighbors across different networks. However, many anchor nodes have different local structures or neighbors across different networks, which could be regarded as anchor nodes’ heterogeneity. It poses a challenge to methods based on the assumption of consistency, as they lack abundant shared information, such as common neighbors. Fortunately, network communities provide the comprehension of node relationships and group structures within networks, which could alleviate the information insufficient. In this article, we propose to address the challenge of inadequate shared information triggered by nodes’ heterogeneity from a community perspective. Our model is based on joint optimization of node representation learning and community discovery, including: 1) a node-level constraint is employed to bring nodes with more anchor pairs as neighbors closer together and 2) a community-level constraint is utilized to bring nodes with higher order similarity closer together. We model the cross-network community alignment relations as asymmetric to mitigate the interference caused by anchor node heterogeneity when measuring community alignment relations. Furthermore, we leverage the learned cross-network community alignment relations to supplement node alignment, which could narrow down the search range of potential anchor nodes by focusing solely on aligning nodes within aligned cross-network communities. We conducted extensive experiments on real-world datasets, and the results show the effectiveness and efficiency of our proposed model on network alignment.

Distributional Learning for Network Alignment with Global Constraints.

Robust Network Alignment with the Combination of Structure and Attribute Embeddings

Geometry interaction network alignment

Attent: Active Attributed Network Alignment

Deep Cross-Network Alignment with Anchor Node Pair Diverse Local Structure.

Aligning Users Across Social Networks Using Network Embedding

Robust Attributed Graph Alignment Via Joint Structure Learning and Optimal Transport

HGENA - A Hyperbolic Graph Embedding Approach for Network Alignment.

Adaptive Network Alignment with Unsupervised and Multi-order Convolutional Networks

Distributed Learning of Predictive Structures from Multiple Tasks over Networks

Structural Representation Learning for User Alignment Across Social Networks

Collaborative Cross-Network Embedding Framework for Network Alignment

Sinkhorn Distance Minimization for Adaptive Semi-Supervised Social Network Alignment

SAlign: A Graph Neural Attention Framework for Aligning Structurally Heterogeneous Networks

Non-translational Alignment for Multi-relational Networks.

Network Alignment Enhanced Via Modeling Heterogeneity of Anchor Nodes

Gromov-Wasserstein Learning for Graph Matching and Node Embedding.

Unsupervised Large-Scale Social Network Alignment via Cross Network Embedding

Enhancing Network Alignment through Multi-Scale Information Fusion

Adversarial Learning for Weakly-Supervised Social Network Alignment.

Alleviate the Impact of Heterogeneity in Network Alignment from Community View