用户可自由切换推荐算法时，数据迁移如何影响各方体验与公平性
Multistakeholder Impacts of Profile Portability in a Recommender Ecosystem
- 打破平台与算法绑定，让用户自主选择推荐算法
- 数据可迁移使小众用户和内容提供者受益更明显
- 政策需平衡用户控制权与算法公平性，避免新不平等
推荐系统优化长期聚焦于算法层面的改进，如多目标模型或重排序。然而，对推荐生态结构变革的研究仍不足。本文探讨算法多元性（即治理文献中的‘中间件’）的影响：将推荐算法与平台解耦，允许用户选择偏好算法。已有模拟研究表明，算法选择有利于小众用户和内容提供者。但这一模式引发关键问题：当用户更换算法时，其数据如何处理？随着多项数据可迁移法规兴起，强化了用户对数据的所有权与控制权。本文分析此类政策对用户建模及各利益相关方结果的影响。研究发现，不同推荐算法在数据可迁移场景下对用户效用的影响存在差异。论文提出关键政策考量，为设计更公平的推荐生态系统提供依据。
原文摘要 · Abstract (English)
Optimizing outcomes for multiple stakeholders in recommender systems has historically focused on algorithmic interventions, such as developing multi-objective models or re-ranking results from existing algorithms. However, structural changes to the recommendation ecosystem itself remain understudied. This paper explores the implications of algorithmic pluralism (also known as "middleware" in the governance literature), in which recommendation algorithms are decoupled from platforms, enabling users to select their preferred algorithm. Prior simulation work demonstrates that algorithmic choice benefits niche consumers and providers. Yet this approach raises critical questions about user modeling in the context of data portability: when users switch algorithms, what happens to their data? Noting that multiple data portability regulations have emerged to strengthen user data ownership and control. We examine how such policies affect user models and stakeholders' outcomes in recommendation setting. Our findings reveal that data portability scenarios produce varying effects on user utility across different recommendation algorithms. We highlight key policy considerations and implications for designing equitable recommendation ecosystems.
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。