Fair and Transparent AI for Hiring: Evaluating Resume-Job Matching, Bias Mitigation, and Human-in-the-Loop Auditing
Open accessAbstract
Automated hiring systems promise efficiency but risk perpetuating bias. We examine resume-job matching using embedding models and fairness audits. We reproduce a CareerBERT-style shared-embedding model trained on the ESCO taxonomy and a large job ads corpus. We compare it to TF-IDF and SBERT baselines, measuring retrieval utility (MRR, Recall@k) on an ESCO/EURES-derived dataset. We then apply IBM’s AIF360 fairness toolkit to evaluate demographic parity, equalized odds, and error-rate metrics across protected groups. We experiment with pre-processing (Reweighing), in-processing (Adversarial Debiasing), and post-processing (Reject Option) interventions. In simulation, we find typical trade-offs: e.g., reweighing can reduce selection-rate disparities by ~10-20 pts at the cost of a few percentage points drop in Recall@10, while adversarial training maintains utility but only partially closes gap. We implement a prototype HR auditing dashboard with group metrics and example rationales. In a human evaluation, auditors preferred the fair-re-ranked shortlist in X% of cases (p<0.05) and reported higher perceived equity. All code, data splits, and figures are released for reproducibility.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Human oversight is not an inherently bias-reducing mechanism; rather, it is contingent upon design or governance decisions.
“The above studies collectively suggest that human oversight is not an inherently bias-reducing mechanism; rather, it is contingent upon design or governance decisions.”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
Section 5.3 describes a deliberately designed audit interface that gives human reviewers fairness metrics, explanations, and flagged cases; the Discussion also calls for decision logging and continuous monitoring (page numbers unavailable). These passages support the importance of design and governance, but the paper does not directly establish that human oversight on its own fails to reduce bias. (LLM, conf=0.78)
Experimental — accuracy not yet measuredThe framework incorporates several checkpoints where human evaluators rate system recommendations for possible bias before decisions are made on candidates.
“The framework also incorporates several checkpoints where human evaluators rate system recommendations for possible bias before decisions are made on candidates.”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
Section 5.3 (page not provided) describes an audit interface that flags possible bias for human review and a study in which HR professionals rated shortlists for perceived fairness. The Discussion says audited outputs should inform, not replace, human judgment. The paper does not describe several checkpoints at which evaluators rate recommendations before candidate decisions are made. (LLM, conf=0.84)
Experimental — accuracy not yet measuredBias can be introduced during various stages of the hiring process, and effective mitigation requires consideration of each of these stages.
“The hierarchical system recognizes that bias can be introduced during various stages of the hiring process, and effective mitigation of this issue requires consideration of each of these stages.”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
The Introduction describes bias in resume screening and evaluates mitigation at pre-processing, in-processing, and post-processing stages. The Conclusion recommends combined strategies and human oversight. The paper does not establish that bias arises at every stage of the broader hiring process or that effective mitigation requires considering each one. (LLM, conf=0.83)
Experimental — accuracy not yet measuredHumans are capable of detecting harms that are not captured by aggregate metrics of models and are able to take corrective actions.
“In such cases, humans are capable of detecting harms that are not captured by aggregate metrics of models and are able to take corrective actions [38,66].”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
Section 7.4 reports that HR raters noticed stereotypical terms in the original model’s recommendations, and the Conclusion says human judgment can uncover subtle biases. The paper does not establish that those harms were missed by aggregate metrics or that the human raters took corrective action; they evaluated shortlists that had already been mitigated. (LLM, conf=0.84)
Experimental — accuracy not yet measuredAlgamaty developed approaches for fair and transparent AI in hiring that combine resume-job matching with bias mitigation and human-in-the-loop auditing.
“Algamaty developed approaches for fair and transparent AI in hiring that combine resume–job matching with bias mitigation and human-in-the-loop auditing [66].”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
The human auditing mechanism is intended not only to pinpoint specific instances of biased recommendations but also to create feedback that can improve system performance over time.
“The human auditing mechanism is intended not only to pinpoint specific instances of biased recommendations but also to create feedback that can improve system performance over time.”
From Human-in-the-Loop Artificial Intelligence: A Systematic Review of Concepts, Methods, and Applications (2026) · cites this paper as [66]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
No references could be resolved yet.
Research works printed but not linked: 6
- Bellamy, R. K. E., Dey, K., Hind, M., Hoffman, S., Houde, S., Kannan, K., Lohia, P., Martino, J., Mehta, S., Mojsilović, A., et al. (2018). AI Fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted algorithmic bias. AAAI/ACM FAccT, 5. (See Fig.1 pipeline).
- Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018). BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv. (Introduced transformer embeddings).
- Fabris, M., Brougham, D., Fotheringham, B., Licht, A., Liu, B., Mines, M., Sokol, K., Xing, B., Zhou, X., Zheng, Y., & Kang, J. (2023). Fairness and Bias in Algorithmic Hiring: A Multidisciplinary Survey. arXiv:2309.13933. (Survey of hiring AI, hazards and opportunities). · arXiv
- Geyik, S. C., Ambler, S., & Kenthapadi, K. (2019). Fairness-Aware Ranking in Search & Recommendation Systems with Application to LinkedIn Talent Search. KDD. (Proposed ranking re-ordering for diversity; deployed at scale with 3× fairness improvement).
- Kaya, M., & Bogers, T. (2025). Mapping Stakeholder Needs to Multi-Sided Fairness in Candidate Recommendation for Algorithmic Hiring. CHI. (Discusses multi-stakeholder fairness; observes much hiring fairness work is conceptual).
- Yu, X., Xu, R., Xue, C., Zhang, J., & Yu, Z. (2025). ConFit v2: Improving Resume-Job Matching using Hypothetical Resume Embedding and Runner-Up Hard-Negative Mining. arXiv:2502.12361. (Introduces LLM-generated "reference resumes" and hard-negative mining; achieves +13.8% Recall and +17.5% nDCG). · arXiv
Cited by
Sorted by citations
Lazaros, K., Vrahatis, A. G., & Kotsiantis, S. (2026) · Entropy · doi:10.3390/e28040377 · Cited by 50 · see citing works
- Authors
- Algamaty, R. K.
- Affiliations
- Libyan Academy (LY)
- Corresponding author
- Ragda Khaled Algamaty
- Venue
- Libyan Open University Journal of Applied Sciences (LOUJAS)
- Publisher
- Libyan Open University
- Published
- 2025-07-01
- Pages
- 01-10
- ISSN
- 3106-0919 (ISSN-L)
- OpenAlex
- W4416819464
- Cited by
- 1
- References
- 0
- Citations by year
- 2026: 1
- FWCI
- 1.05
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- Libyan Open University Journal of Applied Sciences (LOUJAS) (publishedVersion, cc-by)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Social Sciences › Business, Management and Accounting › Organizational Behavior and Human Resource Management › AI and HR Technologies
- Social Sciences › Sociology and Political Science › Names, Identity, and Discrimination Research
- Keywords
- automated hiring, algorithmic bias, fairness auditing, job advertisements, demographic parity, equalized odds, reweighing, adversarial debiasing, reject option, protected groups
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
