[Paper Review] FairGen: Towards Fair Graph Generation

Lecheng Zheng, Dawei Zhou+4 more|arXiv (Cornell University)|Mar 30, 2023

Ethics and Social Impacts of AI4 citations

TL;DR

FairGen is a fairness-aware graph generative model that integrates label information and user-defined parity constraints to mitigate representation disparity in synthetic graphs. By leveraging self-paced learning to progressively train on easy to hard patterns, it generates high-quality graphs that reduce bias against protected groups, improving downstream task performance by up to 17% through data augmentation.

ABSTRACT

There have been tremendous efforts over the past decades dedicated to the generation of realistic graphs in a variety of domains, ranging from social networks to computer networks, from gene regulatory networks to online transaction networks. Despite the remarkable success, the vast majority of these works are unsupervised in nature and are typically trained to minimize the expected graph reconstruction loss, which would result in the representation disparity issue in the generated graphs, i.e., the protected groups (often minorities) contribute less to the objective and thus suffer from systematically higher errors. In this paper, we aim to tailor graph generation to downstream mining tasks by leveraging label information and user-preferred parity constraints. In particular, we start from the investigation of representation disparity in the context of graph generative models. To mitigate the disparity, we propose a fairness-aware graph generative model named FairGen. Our model jointly trains a label-informed graph generation module and a fair representation learning module by progressively learning the behaviors of the protected and unprotected groups, from the `easy' concepts to the `hard' ones. In addition, we propose a generic context sampling strategy for graph generative models, which is proven to be capable of fairly capturing the contextual information of each group with a high probability. Experimental results on seven real-world data sets, including web-based graphs, demonstrate that FairGen (1) obtains performance on par with state-of-the-art graph generative models across nine network properties, (2) mitigates the representation disparity issues in the generated graphs, and (3) substantially boosts the model performance by up to 17% in downstream tasks via data augmentation.

Motivation & Objective

• To address representation disparity in graph generative models, where protected (minority) groups are systematically underrepresented due to unsupervised training objectives.
• To tailor graph generation to downstream mining tasks by incorporating label information and user-preferred fairness constraints.
• To mitigate bias in synthetic graphs by ensuring equitable representation learning for both protected and unprotected groups.
• To develop a generic context sampling strategy that fairly captures group-specific contextual information with high probability.
• To demonstrate that fairness-aware generation improves downstream model performance through data augmentation.

Proposed method

• FairGen employs a self-paced learning paradigm that progressively trains on simpler (eustatic) patterns before moving to complex ones, enabling better capture of protected group behaviors.
• It jointly optimizes a label-informed graph generation module and a fair representation learning module to align synthetic graphs with real-world structural and label distributions.
• A novel context sampling strategy is introduced to ensure high-probability capture of contextual information from both protected and unprotected groups.
• The model uses a reconstruction loss minimized under fairness constraints, ensuring balanced contribution of minority and majority groups to the training objective.
• The framework is trained end-to-end with a dual-objective loss: one for graph structure fidelity and another for fairness in representation learning.
• It leverages node representations from models like Node2Vec for visualization and evaluation, ensuring topological structure preservation across groups.

Experimental results

Research questions

• RQ1 Can fairness constraints be effectively integrated into graph generative models to reduce representation disparity?
• RQ2 How does self-paced learning improve the modeling of underrepresented (protected) groups in graph generation?
• RQ3 To what extent can label-informed generation enhance downstream task performance via data augmentation?
• RQ4 Can a generic context sampling strategy fairly represent both protected and unprotected groups in synthetic graphs?
• RQ5 How does FairGen compare to state-of-the-art graph generators in preserving network properties while ensuring fairness?

Key findings

• FairGen achieves performance on par with state-of-the-art graph generative models across nine network properties on seven real-world datasets.
• The model effectively mitigates representation disparity, as shown by improved topological preservation of protected groups in t-SNE visualizations.
• FairGen boosts downstream model performance by up to 17% through data augmentation, demonstrating its utility in real-world applications.
• The proposed context sampling strategy ensures high-probability capture of group-specific contextual information, enhancing fairness in representation learning.
• Compared to baselines like NetGAN, GAE, and TagGen, FairGen better preserves structural and label-based distinctions between protected and unprotected groups.
• The self-paced learning mechanism enables the model to gradually learn complex patterns from protected groups, reducing systematic bias in generation.