Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-View Graph Consistency Learning for Invariant Graph Representations

Jie Chen, Hua Mao, Wai Lok Woo, Chuanbin Liu, Xi Peng

TL;DR

CGCL tackles invariant graph representation learning for link prediction on incomplete graphs by constructing two complementary augmented graph views and enforcing cross-view consistency between one view and the reconstruction from the other. The method uses a shared GCN encoder and a cross-view decoder to predict missing edges, supported by a coupled augmentation scheme and theoretical analysis of supervisory information. Empirical results on five standard graph datasets show competitive or superior performance against state-of-the-art baselines, along with ablation and sensitivity studies that validate the approach’s robustness. Overall, CGCL provides a principled, self-supervised framework to learn view-invariant representations that improve incomplete-graph reconstruction and link prediction in graph-structured data.

Abstract

Graph representation learning is fundamental for analyzing graph-structured data. Exploring invariant graph representations remains a challenge for most existing graph representation learning methods. In this paper, we propose a cross-view graph consistency learning (CGCL) method that learns invariant graph representations for link prediction. First, two complementary augmented views are derived from an incomplete graph structure through a coupled graph structure augmentation scheme. This augmentation scheme mitigates the potential information loss that is commonly associated with various data augmentation techniques involving raw graph data, such as edge perturbation, node removal, and attribute masking. Second, we propose a CGCL model that can learn invariant graph representations. A cross-view training scheme is proposed to train the proposed CGCL model. This scheme attempts to maximize the consistency information between one augmented view and the graph structure reconstructed from the other augmented view. Furthermore, we offer a comprehensive theoretical CGCL analysis. This paper empirically and experimentally demonstrates the effectiveness of the proposed CGCL method, achieving competitive results on graph datasets in comparisons with several state-of-the-art algorithms.

Cross-View Graph Consistency Learning for Invariant Graph Representations

TL;DR

CGCL tackles invariant graph representation learning for link prediction on incomplete graphs by constructing two complementary augmented graph views and enforcing cross-view consistency between one view and the reconstruction from the other. The method uses a shared GCN encoder and a cross-view decoder to predict missing edges, supported by a coupled augmentation scheme and theoretical analysis of supervisory information. Empirical results on five standard graph datasets show competitive or superior performance against state-of-the-art baselines, along with ablation and sensitivity studies that validate the approach’s robustness. Overall, CGCL provides a principled, self-supervised framework to learn view-invariant representations that improve incomplete-graph reconstruction and link prediction in graph-structured data.

Abstract

Graph representation learning is fundamental for analyzing graph-structured data. Exploring invariant graph representations remains a challenge for most existing graph representation learning methods. In this paper, we propose a cross-view graph consistency learning (CGCL) method that learns invariant graph representations for link prediction. First, two complementary augmented views are derived from an incomplete graph structure through a coupled graph structure augmentation scheme. This augmentation scheme mitigates the potential information loss that is commonly associated with various data augmentation techniques involving raw graph data, such as edge perturbation, node removal, and attribute masking. Second, we propose a CGCL model that can learn invariant graph representations. A cross-view training scheme is proposed to train the proposed CGCL model. This scheme attempts to maximize the consistency information between one augmented view and the graph structure reconstructed from the other augmented view. Furthermore, we offer a comprehensive theoretical CGCL analysis. This paper empirically and experimentally demonstrates the effectiveness of the proposed CGCL method, achieving competitive results on graph datasets in comparisons with several state-of-the-art algorithms.
Paper Structure (23 sections, 15 equations, 4 figures, 2 tables, 1 algorithm)

This paper contains 23 sections, 15 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Cross-View Graph Consistency Learning
  4. Problem Formulation
  5. Network Architecture
  6. CGCL Model
  7. Motivation
  8. Coupled Graph Structure Augmentation
  9. Reconstruction of an Incomplete Graph Structure
  10. Training Scheme
  11. Theoretical Analysis
  12. Related Work
  13. Experiments
  14. Experimental Settings
  15. Datasets
  16. ...and 8 more sections

Figures (4)

  • Figure 1: Framework of the CGCL model. Each augmented view of a graph structure, ${\mathbf{A}_i}$$\left( {i = 1,2} \right)$, corresponds to three modules, including an augmented graph structure module, a shared GCN encoder module and a shared cross-view consistency decoder module. $\mathbf{A}$ and $\widetilde{\mathbf{A}}$ represent the incomplete graph structure and predictive graph structure, respectively.
  • Figure 2: The AUC and AP values yielded by the CGCL method with different combinations of $d_v$ and $r$ on the Cora dataset.
  • Figure 3: The AUC and AP values yielded by the CGCL method with different combinations of $d_v$ and $r$ on the Computers dataset.
  • Figure 4: Convergence results obtained by the CGCL method on all the datasets.