Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mitigating Degree Bias in Signed Graph Neural Networks

Fang He, Jinhai Deng, Ruizhan Xue, Maojun Wang, Zeyu Zhang

TL;DR

This work targets fairness in Signed Graph Neural Networks by identifying degree bias as a key challenge in multi-hop aggregation for graphs with both positive and negative edges. It introduces DD-SGNN, a model-agnostic plugin that performs head-to-tail debiasing through personalized, layer-wise translations to augment tail-node neighborhoods while preserving balance theory semantics, and optimizes a combined loss of fairness and link-sign prediction. Empirical results on four real-world signed networks show that DD-SGNN reduces degree-based disparity (DSP) without sacrificing AUC or F1, with ablations confirming the necessity of each component. The proposed approach provides a generalizable framework to obtain fairer node representations in signed graphs, potentially applicable to a range of SGNN architectures and real-world networks with polarized communities.

Abstract

Like Graph Neural Networks (GNNs), Signed Graph Neural Networks (SGNNs) are also up against fairness issues from source data and typical aggregation method. In this paper, we are pioneering to make the investigation of fairness in SGNNs expanded from GNNs. We identify the issue of degree bias within signed graphs, offering a new perspective on the fairness issues related to SGNNs. To handle the confronted bias issue, inspired by previous work on degree bias, a new Model-Agnostic method is consequently proposed to enhance representation of nodes with different degrees, which named as Degree Debiased Signed Graph Neural Network (DD-SGNN) . More specifically, in each layer, we make a transfer from nodes with high degree to nodes with low degree inside a head-to-tail triplet, which to supplement the underlying domain missing structure of the tail nodes and meanwhile maintain the positive and negative semantics specified by balance theory in signed graphs. We make extensive experiments on four real-world datasets. The result verifies the validity of the model, that is, our model mitigates the degree bias issue without compromising performance($\textit{i.e.}$, AUC, F1). The code is provided in supplementary material.

Mitigating Degree Bias in Signed Graph Neural Networks

TL;DR

This work targets fairness in Signed Graph Neural Networks by identifying degree bias as a key challenge in multi-hop aggregation for graphs with both positive and negative edges. It introduces DD-SGNN, a model-agnostic plugin that performs head-to-tail debiasing through personalized, layer-wise translations to augment tail-node neighborhoods while preserving balance theory semantics, and optimizes a combined loss of fairness and link-sign prediction. Empirical results on four real-world signed networks show that DD-SGNN reduces degree-based disparity (DSP) without sacrificing AUC or F1, with ablations confirming the necessity of each component. The proposed approach provides a generalizable framework to obtain fairer node representations in signed graphs, potentially applicable to a range of SGNN architectures and real-world networks with polarized communities.

Abstract

Like Graph Neural Networks (GNNs), Signed Graph Neural Networks (SGNNs) are also up against fairness issues from source data and typical aggregation method. In this paper, we are pioneering to make the investigation of fairness in SGNNs expanded from GNNs. We identify the issue of degree bias within signed graphs, offering a new perspective on the fairness issues related to SGNNs. To handle the confronted bias issue, inspired by previous work on degree bias, a new Model-Agnostic method is consequently proposed to enhance representation of nodes with different degrees, which named as Degree Debiased Signed Graph Neural Network (DD-SGNN) . More specifically, in each layer, we make a transfer from nodes with high degree to nodes with low degree inside a head-to-tail triplet, which to supplement the underlying domain missing structure of the tail nodes and meanwhile maintain the positive and negative semantics specified by balance theory in signed graphs. We make extensive experiments on four real-world datasets. The result verifies the validity of the model, that is, our model mitigates the degree bias issue without compromising performance(, AUC, F1). The code is provided in supplementary material.
Paper Structure (31 sections, 18 equations, 4 figures, 4 tables)

This paper contains 31 sections, 18 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Signed Graph Neural Networks.
  4. Fairness and Bias in Graph Neural Networks.
  5. Methodology
  6. Preliminaries
  7. Notations.
  8. Task Definition.
  9. Problem Formulation
  10. Degree Bias in SGNN.
  11. How to Define a Fairness Issue in SGNN?
  12. Implementations of DD-SGNN
  13. Aggregation and Propagation in SGNN.
  14. Head-to-Tail Debiasing Method.
  15. Training Constraints and Objective
  16. ...and 16 more sections

Figures (4)

  • Figure 1: Motivation verification on three real-world datasets. Compare downstream task in accuracy metric between edges connected by 'Multiple'-degree nodes and 'Single'-degree nodes across signed graph, using the SGCN as the backbone framework, revealing a significant disparity.
  • Figure 2: The overall framework of DD-SGNN.
  • Figure 3: Ablation study on the effect of each module.
  • Figure 4: DSP performance under different values of hyper-parameter K.