[Paper Review] Fairness in Recommender Systems: Research Landscape and Future Directions

Yashar Deldjoo, Dietmar Jannach+3 more|arXiv (Cornell University)|May 23, 2022

Ethics and Social Impacts of AISocial Sciences18 citations

TL;DR

This survey synthesizes over 160 scholarly works on fairness in recommender systems, identifying key fairness definitions, methodologies, and algorithmic approaches. It reveals that current research often lacks normative grounding, relying on abstract metrics rather than stakeholder-centered fairness, and calls for interdisciplinary research to address real-world fairness challenges.

ABSTRACT

Recommender systems can strongly influence which information we see online, e.g., on social media, and thus impact our beliefs, decisions, and actions. At the same time, these systems can create substantial business value for different stakeholders. Given the growing potential impact of such AI-based systems on individuals, organizations, and society, questions of fairness have gained increased attention in recent years. However, research on fairness in recommender systems is still a developing area. In this survey, we first review the fundamental concepts and notions of fairness that were put forward in the area in the recent past. Afterward, through a review of more than 160 scholarly publications, we present an overview of how research in this field is currently operationalized, e.g., in terms of general research methodology, fairness measures, and algorithmic approaches. Overall, our analysis of recent works points to certain research gaps. In particular, we find that in many research works in computer science, very abstract problem operationalizations are prevalent and questions of the underlying normative claims and what represents a fair recommendation in the context of a given application are often not discussed in depth. These observations call for more interdisciplinary research to address fairness in recommendation in a more comprehensive and impactful manner.

Motivation & Objective

• To map the current research landscape of fairness in recommender systems based on over 160 scholarly publications.
• To analyze how fairness is operationalized in terms of research methodology, fairness measures, and algorithmic design.
• To identify critical research gaps, particularly the lack of normative grounding and stakeholder-centered fairness definitions.
• To highlight the disconnect between algorithmic fairness metrics and real-world ethical implications in recommendation systems.
• To advocate for interdisciplinary research integrating social sciences to address fairness in a more comprehensive and impactful way.

Proposed method

• Systematic review and categorization of 160+ recent publications in computer science focused on fairness in recommender systems.
• Classification of fairness definitions into dimensions such as demographic parity, treatment equality, and exposure fairness.
• Analysis of research methodologies, including real-world data, synthetic data, and A/B testing with logged interactions.
• Evaluation of algorithmic approaches, including fairness-aware optimization, re-ranking, and fairness constraints in model training.
• Application of counterfactual evaluation techniques to assess fairness under intervention scenarios.
• Incorporation of game-theoretic concepts like Walrasian equilibrium to model fairness in reciprocal recommender systems.

Experimental results

Research questions

• RQ1 What are the dominant fairness definitions and dimensions used in current recommender system research?
• RQ2 How is the fairness problem operationalized in terms of methodology, metrics, and algorithmic design across different application domains?
• RQ3 What are the key research gaps in fairness-aware recommendation, particularly regarding normative foundations and stakeholder perception?
• RQ4 How do current fairness metrics align with real-world ethical and societal implications of recommendations?
• RQ5 What role can interdisciplinary collaboration—especially with social sciences—play in advancing fairness in recommender systems?

Key findings

• A majority of fairness research in recommender systems relies on abstract, computationally defined fairness metrics without sufficient grounding in normative or ethical considerations.
• There is a strong reliance on dataset availability for selecting application domains, leading to arbitrary or artificial fairness challenges in many studies.
• The connection between fairness in recommendation and established theories from social sciences, psychology, and ethics is largely absent in current computer science literature.
• Current approaches often treat fairness as an algorithmic optimization problem, neglecting the human and societal dimensions of fairness in real-world contexts.
• Fairness auditing in real platforms remains highly challenging due to technical and access barriers, as illustrated by the case of Facebook NewsFeed being inaccessible to auditors.
• Reciprocal recommender systems, such as in social networks or matching platforms, present a unique fairness challenge due to the need to balance supply, demand, and mutual preferences, which remains under-investigated.