EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Privacy-Preserving Federated Framework of Human Centered System Design for Person-Job Matching incorporating Heterogeneous Knowledge Graphs

Long Xu, Dan Wang, Xinyu Shen, Haitao Zheng, Jiayuan Feng, Zairan Li

Journal of Human-Centered Design for Manufacturing · 01.06.2026 · 2 (2)

DOI: 10.65927/8cz4vg62 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Human-Centered Design for Manufacturing, online erschienen am 01.06.2026. Verfasst von Long Xu und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In talent recruitment and intelligent hiring systems, achieving accurate and explainable human centered system designing for person–job matching and protecting user privacy have become challenges. While most existing approaches rely on centralized data collection or single-source information, which make limitation on their ability to support secure collaboration across organizations and reduces transparency in decision-making. To address these issues, this paper proposes a privacy-preserving federated learning framework for person–job matching that integrates heterogeneous knowledge graphs with explainable recommendation analysis. Using publicly available job–skill datasets, a heterogeneous knowledge graph that captures relationships between jobs and skills, and then learn feature representations through graph neural networks were firstly constructed. Within the federated learning setting, each client stores and trains only on its own local job–skill data, while global collaboration is achieved through parameter aggregation. Differential privacy is further incorporated to minimize the risk of information leakage during model training. To improve interpretability, we introduce skill-contribution decomposition and path-level explanation mechanisms, allowing the framework to highlight the most influential skills and explain how recommendations are generated. Experiments conducted with 20 federated clients demonstrate that the proposed FL-HKG-DP framework consistently outperforms both centralized methods and federated non-graph baselines in terms of Precision@K and NDCG@K. In addition, privacy–utility analysis shows that the model maintains stable performance under practical privacy budgets. The generated explanations also provide clear and auditable evidence for recommendation decisions.Overall, the proposed framework demonstrates strong potential for privacy-compliant recruitment systems and trustworthy intelligent decision-making applications.

Zitieren

APA

Xu, L., Wang, D., Shen, X., Zheng, H., Feng, J., & Li, Z. (2026). A Privacy-Preserving Federated Framework of Human Centered System Design for Person-Job Matching incorporating Heterogeneous Knowledge Graphs. Journal of Human-Centered Design for Manufacturing, 2(2). https://doi.org/10.65927/8cz4vg62

Vancouver

Xu L, Wang D, Shen X, Zheng H, Feng J, Li Z. A Privacy-Preserving Federated Framework of Human Centered System Design for Person-Job Matching incorporating Heterogeneous Knowledge Graphs. Journal of Human-Centered Design for Manufacturing. 2026;2(2). doi:10.65927/8cz4vg62

BibTeX
@article{Xu2026,
  author = {Xu, Long and Wang, Dan and Shen, Xinyu and Zheng, Haitao and Feng, Jiayuan and Li, Zairan},
  title = {A Privacy-Preserving Federated Framework of Human Centered System Design for Person-Job Matching incorporating Heterogeneous Knowledge Graphs},
  journal = {Journal of Human-Centered Design for Manufacturing},
  year = {2026},
  volume = {2},
  number = {2},
  doi = {10.65927/8cz4vg62}
}
Kostenlos starten →