Diffusion recommenders corrected to reduce popularity bias
FairDiff: Mitigating the Self-Reinforcing Matthew Effect in Diffusion Recommender Models
Artificial Intelligence
Summary
Recommender systems suggest items to users, but often popular items get suggested more, making it harder for less popular options to be seen. The authors found that a type of recommender called diffusion models can make this popularity bias even worse by their design. They propose FairDiff, a method that changes how the recommendations are generated to give more chance to less popular, niche items without changing the training process. Their tests show FairDiff can suggest a wider variety of items fairly while keeping good overall recommendations.
What this means in practice
- •For ecommerce platform teams: Improve product recommendations by reducing excessive focus on bestsellers to highlight niche items without retraining models.
- •For music streaming engineers: Generate fairer music recommendations that avoid repeatedly suggesting only popular tracks by adjusting diffusion model generation steps.
Authors
Song-Li Wu, Xianquan Wang, Zhaocheng Du, Weinan Gan, Jingyi Wang
Abstract
While the "Matthew Effect" and filter bubbles are widely recognized outcome-level biases in recommender systems, we reveal that Diffusion Recommender Models (DRMs) uniquely compound this issue through their generative dynamics. Rather than merely inheriting data imbalances, DRMs trigger a self-reinforcing amplification of popularity bias. We identify that this phenomenon is driven by two compounding mechanisms. First, while optimization loss is universally dominated by high-frequency items across recommenders, DRMs suffer from a unique structural prior mismatch during generation. Because the forward terminal distribution of long-tailed data deviates significantly from the standard Gaussian prior, reverse sampling trajectories inherently collapse toward high-density popular items, fundamentally suppressing niche item generation. To dismantle this self-reinforcing loop, we propose FairDiff, a plug-and-play fairness-aware diffusion framework. To overcome the popularity-dominated loss, we introduce Popularity Condition Guidance (PCG). Rather than altering the training objective, PCG acts as an inference-time distributional reweighting mechanism, mathematically reshaping the score-based gradient field to penalize high-popularity regions and guide trajectories toward niche semantics. Furthermore, we design a Semantic Calibration (SC) Module to bridge the prior mismatch, aligning the forward and reverse distributions via one-step optimal transport. Comprehensive evaluations demonstrate that FairDiff achieves state-of-the-art performance while effectively mitigating the self-reinforcing Matthew Effect, highlighting its value as a general framework for DRMs.