Abstract:A real-world graph has a complex topological structure, which is often formed by the interaction of different latent factors. Disentanglement of these latent factors can effectively improve the robustness and expressiveness of the node representation of a graph. However, most existing methods lack consideration of the intrinsic differences in relations between nodes caused by factor entanglement. In this paper, we propose an <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">A dversarial <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">D isentangled <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">G raph <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C onvolutional <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N etwork (ADGCN) for disentangled graph representation learning. To begin with, we point out two aspects of graph disentanglement that need to be considered, i.e., micro-disentanglement and macro-disentanglement. For them, a component-specific aggregation approach is proposed to achieve micro-disentanglement by inferring latent components that caused the links between nodes. On the basis of micro-disentanglement, we further propose a macro-disentanglement adversarial regularizer to improve the separability among component distributions, thus restricting the interdependence among components. Additionally, to reveal the topological graph structure, a diversity-preserving node sampling approach is proposed, by which the graph structure can be progressively refined in a way of local structure awareness. The experimental results on various real-world graph data verify that our ADGCN obtains more favorable performance over currently available alternatives. The source codes of ADGCN are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/SsGood/ADGCN</uri> .

Adversarial Graph Disentanglement With Component-Specific Aggregation.

Adversarial Graph Disentanglement

Multi-level Disentanglement Graph Neural Network

ASA: Adversary Situation Awareness Via Heterogeneous Graph Convolutional Networks.

Disentangled Graph Convolutional Networks

Global Disentangled Graph Convolutional Neural Network Based on a Graph Topological Metric

LGD-GCN: Local and Global Disentangled Graph Convolutional Networks

HSDN: A High-Order Structural Semantic Disentangled Neural Network

Independence Promoted Graph Disentangled Networks

D2-GCN: a graph convolutional network with dynamic disentanglement for node classification

Disentangled Spatiotemporal Graph Generative Models

Graph Meets Probabilistic Generation Model: A New Perspective for Graph Disentanglement

Graph neural networks via contrast between separation and aggregation for self and neighborhood

Disentangled Hyperbolic Representation Learning for Heterogeneous Graphs

Disentangled-based Adversarial Network for Multiplex Network Embedding.

Disentangled Contrastive Learning on Graphs

DGSTD: Learning on Dynamic Graph with Spatio-Temporal Disentanglement

Disentangled contrastive learning for fair graph representations

DisenKGAT: Knowledge Graph Embedding with Disentangled Graph Attention Network

Disentangled Relational Graph Neural Network with Contrastive Learning for Knowledge Graph Completion

Adversarial Training for Graph Neural Networks via Graph Subspace Energy Optimization