Normative Alignment of Recommender Systems via Internal Label Shift

2026-07-12

AI Summary

This work addresses the challenge that recommendation systems, when optimized solely for user engagement, often fail to satisfy normative objectives such as fairness and diversity. To bridge this gap, the authors propose NAILS, a novel method that introduces internal label shift into the normative alignment of recommender systems. Leveraging a hierarchical classification framework, NAILS jointly adjusts both the user-conditional distribution and attribute-level marginal distributions without requiring model retraining, thereby aligning recommendations with predefined normative targets. Experimental results across multiple datasets demonstrate that NAILS significantly enhances attribute-level normative alignment while exerting minimal impact on user engagement, offering an efficient, scalable, and retraining-free alignment mechanism.

Abstract

We introduce NAILS (Normative Alignment of Recommender Systems via Internal Label Shift), a simple and scalable method for aligning recommendation outputs with target distributions over item-level attributes, such as categories. Recommender systems optimized solely for user engagement often fail to satisfy broader normative objectives, including fairness, diversity, and editorial values. NAILS modifies the user-conditional item distribution to induce a specified marginal distribution over attributes while preserving the preferences learned by an existing recommender system and requiring no model retraining. We formulate this problem as a form of label shift applied internally within a hierarchical classification framework. By adopting a stakeholder-centric perspective, NAILS enables recommendation outputs to be aligned with global normative objectives. Empirically, we show that NAILS consistently improves attribute-level alignment with minimal impact on user engagement, providing a practical mechanism for value-driven recommendation.

Problem

• recommender systems
• normative alignment
• fairness
• diversity
• label shift

Innovation

• Normative Alignment
• Internal Label Shift
• Recommender Systems
• Attribute Distribution
• Fairness and Diversity