FSGR: Mitigating Token Frequency Bias for Fair SID-Based Generative Recommendation

2026-08-13

AI Summary

This work addresses the issue of exposure unfairness in semantic ID (SID)-based generative recommendation, where token frequency bias leads to over-prediction of high-frequency IDs and underestimation of low-frequency ones. The study systematically identifies this problem for the first time and proposes the Fair Semantic Generative Recommendation (FSGR) framework. FSGR balances the semantic representation space during SID construction via optimal transport allocation and a dual-criterion re-anchoring mechanism, and further refines fairness during recommendation training through a two-stage strategy incorporating hierarchical frequency calibration. By leveraging the hierarchical semantic structure of SIDs, the approach jointly optimizes representation balance and downstream fairness. Experiments across three public datasets and three backbone models demonstrate an average improvement of over 20% in Gini-based fairness while maintaining competitive recommendation accuracy.

Abstract

Semantic ID (SID)-based generative recommendation has recently achieved remarkable success. However, existing methods suffer from a previously overlooked fairness issue, which we term \textbf{Token Frequency Bias}, where high-frequency SID tokens are systematically over-predicted while low-frequency SID tokens are under-predicted. This bias originates from the combined effects of imbalanced semantic codebooks during SID construction, and popularity bias together with the maximum likelihood estimation objective during recommendation training, resulting in unfair exposure across item categories. Existing SID methods mainly focus on improving codebook quality and overlook the impact of token frequency imbalance on downstream recommendation fairness, while LLM debiasing methods often yield suboptimal results when directly applied to SID-based recommendation, due to the hierarchical semantics of SID tokens. To address this issue, we propose \textbf{FSGR}, a fairness optimization framework for SID-based generative recommendation. During SID construction, FSGR employs OT-based Assignment Optimization and Dual-Criteria Re-anchor mechanism to form a more balanced SID representation space. During recommendation training, it adopts a two-stage training strategy and introduces Hierarchical Frequency Calibration for layer-specific fairness fine-tuning. Experiments on three public datasets with three backbone models demonstrate that FSGR mitigates token frequency bias and delivers an average Gini fairness improvement of over 20\% while maintaining competitive recommendation accuracy.

Problem

• Token Frequency Bias
• Fairness
• Generative Recommendation
• Semantic ID
• Exposure Fairness

Innovation

• Token Frequency Bias
• Semantic ID
• Fairness Optimization
• Optimal Transport
• Hierarchical Frequency Calibration