Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DropEdge not Foolproof: Effective Augmentation Method for Signed Graph Neural Networks

Zeyu Zhang, Lu Li, Shuyan Wan, Sijie Wang, Zhiyi Wang, Zhiyuan Lu, Dong Hao, Wanli Li

TL;DR

The authors introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training.

Abstract

The paper discusses signed graphs, which model friendly or antagonistic relationships using edges marked with positive or negative signs, focusing on the task of link sign prediction. While Signed Graph Neural Networks (SGNNs) have advanced, they face challenges like graph sparsity and unbalanced triangles. The authors propose using data augmentation (DA) techniques to address these issues, although many existing methods are not suitable for signed graphs due to a lack of side information. They highlight that the random DropEdge method, a rare DA approach applicable to signed graphs, does not enhance link sign prediction performance. In response, they introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training. Experimental results show that SGA significantly boosts the performance of SGNN models, with a notable 32.3% improvement in F1-micro for SGCN on the Slashdot dataset.

DropEdge not Foolproof: Effective Augmentation Method for Signed Graph Neural Networks

TL;DR

The authors introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training.

Abstract

The paper discusses signed graphs, which model friendly or antagonistic relationships using edges marked with positive or negative signs, focusing on the task of link sign prediction. While Signed Graph Neural Networks (SGNNs) have advanced, they face challenges like graph sparsity and unbalanced triangles. The authors propose using data augmentation (DA) techniques to address these issues, although many existing methods are not suitable for signed graphs due to a lack of side information. They highlight that the random DropEdge method, a rare DA approach applicable to signed graphs, does not enhance link sign prediction performance. In response, they introduce the Signed Graph Augmentation (SGA) framework, which includes a structure augmentation module to identify candidate edges and a strategy for selecting beneficial candidates, ultimately improving SGNN training. Experimental results show that SGA significantly boosts the performance of SGNN models, with a notable 32.3% improvement in F1-micro for SGCN on the Slashdot dataset.
Paper Structure (27 sections, 1 theorem, 25 equations, 12 figures, 6 tables, 2 algorithms)

This paper contains 27 sections, 1 theorem, 25 equations, 12 figures, 6 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Proposed Method
  4. Generating Candidate Training Samples
  5. Selecting Beneficial Candidate Training Samples
  6. Introducing New Feature for Training Samples
  7. Generalization Bound of SGNN
  8. Experiments
  9. Performance Evaluation (Q1)
  10. Ablation Study (Q2)
  11. Parameter Sensitivity Analysis (Q3)
  12. Conclusion
  13. Experimental results of data augmentation through random structural perturbations
  14. Related Work
  15. Signed Graph Neural Networks
  16. ...and 12 more sections

Key Result

Theorem 1

Figures (12)

  • Figure 1: Green and red lines represent positive and negative edges, resp. Solid lines represent edges in the training set, while dashed lines represent edges in the test set.
  • Figure 2: Effectiveness of random EdgeDrop (SGCN derr2018signed as backbone model) on link sign prediction performance with six real-world signed graph datasets.
  • Figure 3: The overall process of SGA. Green lines represent positive edges and red lines represent negative edges.
  • Figure 4: Case Study of SGA. Note that green lines denote positive edges and red lines denote negative edges.
  • Figure 5: Performance of SGCN: AUC scores (with standard deviation) across six benchmark datasets, evaluated under variations in parameters $\epsilon_{del}^+$, $\epsilon_{del}^-$, $\epsilon_{add}^+$, $\epsilon_{add}^-$, $T$ and $\lambda_0$.
  • ...and 7 more figures

Theorems & Definitions (4)

  • Definition 3.1
  • Definition 3.2: Local Balance Degree
  • Definition 3.3: Edge Difficulty Score
  • Theorem 1: Generalization Gap of SGNN