III: Medium: Collaborative Research: Fair Recommendation Through Social Choice

Recommender systems are machine learning systems that provide personalized access to information, media and e-commerce catalogs. These systems are widely used and are central to Americans' experience of the Internet. However, concern has grown that these systems can have negative impacts on both individuals and society more generally, by propagating biases, excluding minoritized sub-groups from recommendation results, and offering less optimal performance to individuals with non-mainstream viewpoints.

These issues, as well as other potential harms, have been the topic of recent research attention. However, the practical success of this work has been limited because fairness has generally been conceived in simple, narrow ways, e.g. fairness relative to a single group, and because it has remained largely divorced from real-world organizational practices.

In this research, the investigators will overcome both of these limitations. They will conduct a detailed contextual analysis of fairness within a non-profit organization, ensuring that their fairness concepts are grounded in real organizational needs. The ensuing implementation of fair recommendation will reflect the complexities of practice by representing and balancing the viewpoints of different stakeholders.

The work will enhance our understanding of algorithmic fairness as a situated and complex concept and of the development challenges arising throughout the full life-cycle of fair machine learning.

The multidisciplinary team on this project includes experts in recommender systems, computational social choice, and philanthropic informatics. The team will create new fairness-aware recommendation algorithms that are fundamentally multi-agent in nature and based on algorithmic game theory. From this novel vantage point, the project will reformulate recommendation fairness as a combination of social choice allocation and aggregation problems, which integrate both fairness concerns and personalized recommendation provisions, and derive new recommendation techniques based on this formulation.

Working with their non-profit partner, the researchers will conduct interviews and focus groups with diverse stakeholders, building models of the different ways that fairness is operationalized within this organizational context, and generalize these techniques to apply to other organizations. The project will create a model deployment of their multi-stakeholder fairness solution and use both quantitative and qualitative techniques to evaluate it from the perspective of both users and internal stakeholders.

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.