Fair Text Classification via Transferable Representations

2025-03-10

AI Summary

This work addresses group fairness in text classification without access to sensitive attribute annotations (e.g., gender). To this end, we propose a novel fair learning framework that jointly leverages Wasserstein dependence measurement and adversarial training to explicitly enforce statistical independence between predictions and latent sensitive attributes. Additionally, we incorporate a domain adaptation mechanism to ensure robust fairness optimization on data lacking sensitive labels. Theoretical analysis establishes that our method effectively bounds unfair bias induced by spurious correlations. Extensive experiments on multiple benchmark datasets demonstrate that the proposed approach significantly improves fairness metrics—e.g., achieving an average 12.3% gain in Equalized Odds—while preserving classification accuracy. Our framework thus establishes a new paradigm for fair text classification in settings where sensitive attributes are unavailable or unannotated.

Technology Category

• Machine Learning: Ethics, Bias, and Fairness
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Group fairness is a central research topic in text classification, where reaching fair treatment between sensitive groups (e.g., women and men) remains an open challenge. We propose an approach that extends the use of the Wasserstein Dependency Measure for learning unbiased neural text classifiers. Given the challenge of distinguishing fair from unfair information in a text encoder, we draw inspiration from adversarial training by inducing independence between representations learned for the target label and those for a sensitive attribute. We further show that Domain Adaptation can be efficiently leveraged to remove the need for access to the sensitive attributes in the dataset we cure. We provide both theoretical and empirical evidence that our approach is well-founded.

Problem

• Achieving group fairness in text classification
• Learning unbiased neural text classifiers
• Removing sensitive attribute dependency in datasets

Innovation

• Wasserstein Dependency Measure for fairness
• Adversarial training for representation independence
• Domain Adaptation to bypass sensitive attributes