III: Medium: Towards Inclusive Recommendation Systems with Stakeholder Alignment

随着推荐系统在日常生活的各个方面不断影响不同的利益相关者，适应他们的不同目标是至关重要的。传统的推荐方法只关注优化准确性和相关指标，而忽略了其他不同的利益相关者相关目标。该项目代表了将不同利益相关者的目标整合到推荐系统的设计和部署中的系统性努力。这包括描述他们的目标，设计包含和优化不同目标的推荐方法，改进数据质量，以及理解不希望的系统行为的驱动因素。这一全面努力符合推进可信赖人工智能的国家利益，并将包括竞赛、研讨会和互动演示等外展活动。该项目旨在研究设计与多个利益相关者的目标一致的推荐系统所需的基本组件。为了实现这一目标，研究将(i)开发框架，使用数据作为软度量来捕获复杂的、依赖于上下文的目标；（ii）设计方法以改善数据质素，并协助与不同目标保持一致；（iii）开发博弈论推荐算法，以实现所有利益相关者都能接受的不同目标之间的权衡；（iv）开发框架，将系统范围内的行为归因于向推荐系统提供数据的个人。通过这些努力，该项目显著扩展了机器学习系统对齐机制设计的基础知识，拓宽了我们对多利益相关者环境中数据影响的理解。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

As recommender systems continue to impact diverse stakeholders in various aspects of daily life, accommodating their distinct objectives is crucial. Traditional recommendation methodologies have focused solely on optimizing accuracy and related metrics, neglecting other diverse stakeholder-dependent objectives. This project represents a systematic effort to incorporate the objectives of different stakeholders into the design and deployment of a recommender system. This includes characterizing their objectives, designing recommendation approaches that incorporate and optimize different objectives, improving data quality, and understanding the drivers of undesired system behaviors. This comprehensive effort serves the national interest of advancing trustworthy AI and will include outreach initiatives such as competitions, workshops, and interactive demonstrations. This project aims to investigate the fundamental components necessary for designing a recommender system that aligns with the objectives of multiple stakeholders. To achieve this, the research will (i) develop frameworks that use data as a soft metric to capture complex, context-dependent objectives; (ii) design methods to improve data quality and facilitate the alignment with different objectives; (iii) develop game-theoretic recommendation algorithms to achieve a tradeoff in different objectives that is acceptable to all stakeholders; and (iv) develop frameworks that attribute system-wide behaviors to individuals who provide data to a recommender system. Through these efforts, this project significantly expands the foundational knowledge of alignment mechanism design for machine learning systems and broadens our understanding of the impact of data in a multi-stakeholder environment.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.