Endowing Pre-trained Graph Models with Provable Fairness

Zhongjian Zhang, Mengmei Zhang, Yue Yu, Cheng Yang, Jiawei Liu, Chuan Shi

Keywords: Graph Neural Networks, Fairness, Pre-trained Graph Models

Abstract: Pre-trained graph models (PGMs) have received considerable attention in graph machine learning by capturing transferable inherent structural properties and applying them to different downstream tasks. Similar to pre-trained language models, PGMs also inherit biases from human society, resulting in discriminatory behavior in downstream applications. However, the debiasing process of most existing methods is coupled with parameter optimization of GNN, making them not efficient to debias PGMs. Moreover, these debiasing methods lack a theoretical guarantee, i.e., provable lower bounds on the fairness of model predictions, which directly provides assurance in a practical scenario. To overcome these limitations, we propose a novel framework that endows pre-trained graph models with provable fairness (called GraphPAR). GraphPAR freezes the parameters of PGMs and applies a parameter-efficient adapter on node representations to make the model's predictions fairer. Specifically, we design a sensitive attribute augmenter that extends node representations with different sensitive attribute semantics for each node. Then employ two adversarial debiasing methods to optimize the adapter's parameters. Furthermore, based on the proposed framework GraphPAR, we quantify whether the fairness of each node is provable fairness, i.e., predictions are always fair within a certain range of sensitive attribute semantics. Experimental evaluations on real-world datasets demonstrate that GraphPAR achieves state-of-the-art performance and fairness on node classification task. Furthermore, based on our GraphPAR, around 90% nodes have provable fairness.

Track: Graph Algorithms and Learning for the Web

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: Yes

Submission Number: 2385