Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Fair Graph Representation Learning in Social Networks

Guixian Zhang, Guan Yuan, Debo Cheng, Lin Liu, Jiuyong Li, Shichao Zhang

TL;DR

This work identifies that a fundamental reason for the unfairness of GNNs in social network learning is the phenomenon of social homophily, i.e., users in the same group are more inclined to congregate, and proposes a method called Equity-Aware GNN (EAGNN) towards fair graph representation learning.

Abstract

With the widespread use of Graph Neural Networks (GNNs) for representation learning from network data, the fairness of GNN models has raised great attention lately. Fair GNNs aim to ensure that node representations can be accurately classified, but not easily associated with a specific group. Existing advanced approaches essentially enhance the generalisation of node representation in combination with data augmentation strategy, and do not directly impose constraints on the fairness of GNNs. In this work, we identify that a fundamental reason for the unfairness of GNNs in social network learning is the phenomenon of social homophily, i.e., users in the same group are more inclined to congregate. The message-passing mechanism of GNNs can cause users in the same group to have similar representations due to social homophily, leading model predictions to establish spurious correlations with sensitive attributes. Inspired by this reason, we propose a method called Equity-Aware GNN (EAGNN) towards fair graph representation learning. Specifically, to ensure that model predictions are independent of sensitive attributes while maintaining prediction performance, we introduce constraints for fair representation learning based on three principles: sufficiency, independence, and separation. We theoretically demonstrate that our EAGNN method can effectively achieve group fairness. Extensive experiments on three datasets with varying levels of social homophily illustrate that our EAGNN method achieves the state-of-the-art performance across two fairness metrics and offers competitive effectiveness.

Towards Fair Graph Representation Learning in Social Networks

TL;DR

This work identifies that a fundamental reason for the unfairness of GNNs in social network learning is the phenomenon of social homophily, i.e., users in the same group are more inclined to congregate, and proposes a method called Equity-Aware GNN (EAGNN) towards fair graph representation learning.

Abstract

With the widespread use of Graph Neural Networks (GNNs) for representation learning from network data, the fairness of GNN models has raised great attention lately. Fair GNNs aim to ensure that node representations can be accurately classified, but not easily associated with a specific group. Existing advanced approaches essentially enhance the generalisation of node representation in combination with data augmentation strategy, and do not directly impose constraints on the fairness of GNNs. In this work, we identify that a fundamental reason for the unfairness of GNNs in social network learning is the phenomenon of social homophily, i.e., users in the same group are more inclined to congregate. The message-passing mechanism of GNNs can cause users in the same group to have similar representations due to social homophily, leading model predictions to establish spurious correlations with sensitive attributes. Inspired by this reason, we propose a method called Equity-Aware GNN (EAGNN) towards fair graph representation learning. Specifically, to ensure that model predictions are independent of sensitive attributes while maintaining prediction performance, we introduce constraints for fair representation learning based on three principles: sufficiency, independence, and separation. We theoretically demonstrate that our EAGNN method can effectively achieve group fairness. Extensive experiments on three datasets with varying levels of social homophily illustrate that our EAGNN method achieves the state-of-the-art performance across two fairness metrics and offers competitive effectiveness.

Paper Structure

This paper contains 28 sections, 6 theorems, 24 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Notations
  4. Causal View
  5. Social Homophily
  6. Fairness Metric
  7. The Proposed EAGNN Method
  8. Encoding and Classification
  9. Sufficiency
  10. Independence
  11. Separation
  12. Model Training
  13. Experiments
  14. Experiment setup
  15. Real-world datasets
  16. ...and 13 more sections

Key Result

Theorem 1

Let $\mathcal{G}$ be a graph defined by ${\mathcal{V}, \mathcal{E}}$. Each node $v_i$ in $\mathcal{G}$ is characterised by a feature vector $\mathbf{x}_i \in \mathbb{R}^l$ and a sensitive attribute $s_i$. For any node $v_i \in \mathcal{V}$ belonging to group $b$, the expectation of the pre-activatio where $\mathbf{W}$ is the parameter matrix in the GCN and $\mathcal{D}_{s_{i}}$ is the neighbour di

Figures (5)

  • Figure 1: An example of graph data is provided to illustrate the social homophily. $S$ denotes the sensitive attribute, $Y$ denotes the label, and $\hat{Y}$ denotes the prediction.
  • Figure 2: The proposed SCM for representing the graph data generation process. We aim to avoid building spurious correlations between $S$ and $Y$ during the training process.
  • Figure 3: Sensitivity analysis for the $\mathcal{L}_{suff}$ on Credit and German.
  • Figure 4: Sensitivity analysis for the $\mathcal{L}_{in}$ and $\mathcal{L}_{se}$ on three real-world datasets.
  • Figure 5: Sensitivity analysis for $\mathcal{L}_{suff}$ on Bail.

Theorems & Definitions (9)

  • Definition 1: Social Homophily
  • Definition 2: Statistical Parity dwork2012fairness
  • Definition 3: Equal Opportunity hardt2016equality
  • Theorem 1
  • Theorem 2
  • Theorem 3
  • Theorem 1
  • Theorem 2
  • Theorem 3