[Paper Review] ProFairRec: Provider Fairness-aware News Recommendation

Tao Qi, Fangzhao Wu+6 more|arXiv (Cornell University)|Apr 10, 2022

Recommender Systems and Techniques4 citations

TL;DR

This paper proposes ProFairRec, a provider fairness-aware news recommendation framework that learns bias-independent representations for news and users to mitigate unfairness toward minority news providers. By decomposing provider bias into separate fair and biased representations, applying adversarial learning and orthogonal regularization, ProFairRec improves fairness without sacrificing recommendation accuracy on public benchmarks.

ABSTRACT

News recommendation aims to help online news platform users find their preferred news articles. Existing news recommendation methods usually learn models from historical user behaviors on news. However, these behaviors are usually biased on news providers. Models trained on biased user data may capture and even amplify the biases on news providers, and are unfair for some minority news providers. In this paper, we propose a provider fairness-aware news recommendation framework (named ProFairRec), which can learn news recommendation models fair for different news providers from biased user data. The core idea of ProFairRec is to learn provider-fair news representations and provider-fair user representations to achieve provider fairness. To learn provider-fair representations from biased data, we employ provider-biased representations to inherit provider bias from data. Provider-fair and -biased news representations are learned from news content and provider IDs respectively, which are further aggregated to build fair and biased user representations based on user click history. All of these representations are used in model training while only fair representations are used for user-news matching to achieve fair news recommendation. Besides, we propose an adversarial learning task on news provider discrimination to prevent provider-fair news representation from encoding provider bias. We also propose an orthogonal regularization on provider-fair and -biased representations to better reduce provider bias in provider-fair representations. Moreover, ProFairRec is a general framework and can be applied to different news recommendation methods. Extensive experiments on a public dataset verify that our ProFairRec approach can effectively improve the provider fairness of many existing methods and meanwhile maintain their recommendation accuracy.

Motivation & Objective

• To address provider unfairness in news recommendation caused by biased user behavior data.
• To develop a general framework that improves fairness without compromising recommendation accuracy.
• To learn provider-fair representations that are independent of news provider bias while preserving user and news preferences.
• To ensure fair exposure for minority news providers through representation disentanglement and adversarial regularization.
• To enable integration with existing recommendation models to enhance fairness across diverse architectures.

Proposed method

• Decompose news representations into provider-fair and provider-biased components using content and provider ID inputs.
• Learn provider-fair and -biased user representations by aggregating user click history with both fair and biased representations.
• Use only provider-fair representations for user-news matching during inference to ensure fairness.
• Apply adversarial learning to make provider-fair news representations indistinguishable across different providers, reducing bias encoding.
• Enforce orthogonality between provider-fair and -biased representations via regularization to improve disentanglement.
• Integrate the framework with existing recommendation models as a plug-in solution, maintaining model-agnostic flexibility.

Experimental results

Research questions

• RQ1 Can a general framework improve provider fairness in news recommendation without degrading recommendation accuracy?
• RQ2 How effective is representation disentanglement in separating bias-independent signals from provider bias?
• RQ3 To what extent does adversarial learning reduce provider bias in fair representations?
• RQ4 How does orthogonal regularization contribute to bias reduction in fair representations?
• RQ5 What trade-off exists between fairness and recommendation performance when tuning adversarial loss weights?

Key findings

• ProFairRec significantly improves provider fairness across multiple SOTA news recommendation models on the MIND dataset.
• The ablation study confirms that removing provider-biased representations harms fairness, proving their role in disentanglement.
• Adversarial learning enhances fairness by reducing provider information in fair representations, with performance improving as adversarial loss weight increases.
• Orthogonal regularization improves fairness by better separating bias and non-bias signals in representations.
• A trade-off exists between fairness and accuracy: higher adversarial loss weights improve fairness but reduce recommendation performance.
• ProFairRec maintains high recommendation accuracy while achieving substantial fairness gains, demonstrating generalization across different recommendation architectures.