Echoes in Filter Bubble: Diagnosing and Curing Popularity Bias in Generative Recommenders

2026-05-16

AI Summary

Generative recommender systems exacerbate popularity bias due to suboptimal token-level optimization and undifferentiated item tokenization, leading to unfair and less diverse recommendations. To address this, this work proposes Ghost, a novel framework that jointly mitigates bias through asymmetric negative likelihood optimization and a skeleton-based semantic item tokenization mechanism, operating synergistically on both the optimization objective and representation structure. Experimental results demonstrate that Ghost significantly alleviates popularity bias across three benchmark datasets, achieving markedly improved fairness and diversity in recommendations while incurring only marginal degradation in overall recommendation performance.

Abstract

Recently, Generative Recommenders (GRs), characterized by a unified end-to-end framework, have exhibited astonishing potential in transforming the recommendation paradigm. Despite their effectiveness, we recognize that GRs are still susceptible to the long-standing issue of popularity bias that has pervaded the recommendation community. Although a few studies have attempted to extend traditional debiasing methods to GRs, their effectiveness is marginal, and the fundamental reason why GRs suffer from popularity bias remains under-explored. To bridge this gap, this study focuses on two core aspects in GRs: the optimization of generative framework and the item tokenization based on semantic index. Based on theoretical analyses, we identify that the severe popularity bias emerges from the confluence of a token-level optimization flaw and the undifferentiated property of item tokenization. Accordingly, this study develops a novel generative recommender system, called Ghost, by designing the asymmetric unlikelihood optimization and the skeleton-founded tokenization. Extensive empirical evaluations across three datasets, alongside multiple SOTA baselines, reveal that Ghost substantially alleviates popularity bias and promotes fairer recommendations, while incurring slight degradation to the overall recommendation utility.

Problem

• popularity bias
• generative recommenders
• recommendation fairness
• tokenization
• optimization flaw

Innovation

• Generative Recommenders
• Popularity Bias
• Asymmetric Unlikelihood Optimization
• Skeleton-Founded Tokenization
• Fair Recommendation