系统梳理推荐系统中缓解反馈循环偏见的前沿方法
Bias Mitigation for AI-Feedback Loops in Recommender Systems: A Systematic Literature Review and Taxonomy
- 从24篇论文中提炼出应对AI反馈循环的六维分类体系
- 发现仅6项研究同时评估公平性与性能，多数方法缺乏长期验证
- 为从业者提供可落地的选型清单，为研究者指明关键缺口
推荐系统持续基于用户对其预测的反馈进行再训练，形成加剧偏见的AI反馈循环，长期会削弱公平性。尽管此风险已广为人知，大多数偏见缓解方法仅在静态数据集上测试，其在多轮再训练中的长期公平性尚不明确。本文对2019-2025年间24篇明确考虑反馈循环并经多轮仿真或真实A/B测试验证的研究进行系统综述。每篇论文按六维度编码：缓解技术、所针对偏见、动态测试设置、评估重点、应用领域及机器学习任务，构建可复用的分类体系。该分类体系为产业界提供快速选型检查表，为学术界指明最紧迫的研究空白，如共享模拟器缺失、评估指标不统一，以及多数研究仅报告公平性或性能，仅有6项同时兼顾两者。
原文摘要 · Abstract (English)
Recommender systems continually retrain on user reactions to their own predictions, creating AI feedback loops that amplify biases and diminish fairness over time. Despite this well-known risk, most bias mitigation techniques are tested only on static splits, so their long-term fairness across multiple retraining rounds remains unclear. We therefore present a systematic literature review of bias mitigation methods that explicitly consider AI feedback loops and are validated in multi-round simulations or live A/B tests. Screening 347 papers yields 24 primary studies published between 2019-2025. Each study is coded on six dimensions: mitigation technique, biases addressed, dynamic testing set-up, evaluation focus, application domain, and ML task, organising them into a reusable taxonomy. The taxonomy offers industry practitioners a quick checklist for selecting robust methods and gives researchers a clear roadmap to the field's most urgent gaps. Examples include the shortage of shared simulators, varying evaluation metrics, and the fact that most studies report either fairness or performance; only six use both.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。