Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
Mapping Stakeholder Needs to Multi-Sided Fairness in Candidate Recommendation for Algorithmic Hiring
Citation. Kaya, M., & Bogers, T. (2025). Mapping Stakeholder Needs to Multi-Sided Fairness in Candidate Recommendation for Algorithmic Hiring. In RecSys ‘25: Proceedings of the Nineteenth ACM Conference on Recommender Systems (pp. 257–267). Association for Computing Machinery. https://doi.org/10.1145/3705328.3748079
Summary. This peer-reviewed RecSys paper uses semi-structured interviews with 40 job seekers, company representatives, recruiters, and recruitment-agency employees to map lived experiences and definitions of fairness in a Danish human-in-the-loop candidate-recommendation system to fairness-metric families. It argues that hiring recommender systems should assess fairness across job seekers, companies, recruiters, and the agency rather than optimizing only one side. The ingested Drive PDF is arXiv v1; the canonical DOI identifies the later published conference version.
Key claims
- Across stakeholder groups, initial fairness definitions primarily reflected individual fairness centered on qualifications, but discussion of candidate lists and evaluation shifted attention toward group fairness for job seekers’ gender and age and companies’ size.
- Stakeholder priorities mapped to Conditional Demographic (Dis)Parity, rank-aware metrics such as skew, diversity, coverage, and company-focused parity measures rather than a single universal fairness metric.
- Fairness assessment should extend beyond shortlisted or contacted candidates to applicants and people not shortlisted, while recognizing that qualification criteria can themselves reproduce historical inequality.
Evidence & limitations
- The study conducted 40 semi-structured interviews in one Danish recruitment-agency context: 9 job seekers, 5 companies, 10 recruiters, and 16 other agency employees; findings are qualitative and context-specific.
- The authors report limited diversity among job-seeker and company participants and did not empirically evaluate hiring outcomes with the proposed metrics.
- The Drive PDF was arXiv:2508.00908v1, while public arXiv metadata and the IT University of Copenhagen publication record link the title to the published RecSys 2025 DOI; the Drive binary hash is retained above and no binary identity with the published version is claimed.