Graph Neural Network Surrogates of Fair Graph Filtering

TL;DR

This paper introduces a graph neural network framework that approximates graph filters, ensuring fairness constraints like statistical parity are met with minimal impact on original scores across various tasks.

Contribution

It presents a universal approximation approach for fair graph filtering using GNNs, enabling runtime optimization of fairness objectives with minimal utility loss.

Findings

Outperforms or matches existing methods in fairness constraints

Maintains high AUC in community recommendation tasks

Creates minimal utility loss in diffusion processes

Abstract

Graph filters that transform prior node values to posterior scores via edge propagation often support graph mining tasks affecting humans, such as recommendation and ranking. Thus, it is important to make them fair in terms of satisfying statistical parity constraints between groups of nodes (e.g., distribute score mass between genders proportionally to their representation). To achieve this while minimally perturbing the original posteriors, we introduce a filter-aware universal approximation framework for posterior objectives. This defines appropriate graph neural networks trained at runtime to be similar to filters but also locally optimize a large class of objectives, including fairness-aware ones. Experiments on a collection of 8 filters and 5 graphs show that our approach performs equally well or better than alternatives in meeting parity constraints while preserving the AUC of score-based community member recommendation and creating minimal utility loss in prior diffusion.