FSGR: A New Method to Mitigate Token Frequency Bias in SID-Based Generative Recommendation

Researchers have introduced FSGR, a novel approach to address token frequency bias in Semantic ID (SID)-based generative recommendation systems. The bias, which leads to unfair exposure across item categories, arises from imbalanced semantic codebooks and popularity bias during training. FSGR aims to correct this by rebalancing token frequencies, improving fairness without sacrificing performance. The method is detailed in a paper on arXiv (2608.12845v1), submitted as a cross-type announcement. The work addresses a critical gap in existing SID methods, which focus on codebook quality but overlook downstream fairness, and LLM debiasing techniques that are suboptimal for SID's hierarchical semantics. FSGR's approach could enhance recommendation systems across various platforms, ensuring more equitable item visibility.