Dynamic labor market matching via deep reinforcement learning: continuous skill taxonomy evolution in technological transitions

Yingying Cen, Haytham F. Isleem and Ghanshyam G. Tejani

International Journal of Manpower, 2025, vol. 47, issue 2, 193-213

Abstract: PurposeRapid digitalization, automation and AI are widening skill mismatches and slowing hiring, and depressing productivity overall. We propose and evaluate an adaptive labor-market matching framework that treats the skill taxonomy as a living, reinforcement-learning-controlled graph, continuously updated from labor-market signals to improve matching quality and time-to-hire.Design/methodology/approachThe system integrates labor-economics mechanisms with modern AI. A multilingual transformer pipeline extracts skills from unstructured worker profiles and job ads; a graph neural network evolves the taxonomy; and a multi-agent reinforcement-learning layer optimizes matching policies while a feedback module aligns ranking signals with long-horizon outcomes (productivity and tenure). Evidence combines a longitudinal corpus spanning 2010–2025 across 18 countries (1.2 M worker profiles; 8.7 M job postings) with stress-test simulations against shocks and emerging skills.FindingsThe framework improves Precision@1 by 9.4% over the best baseline, reduces time-to-hire by 37.2% versus static taxonomies (16.5% versus the best learning baseline), and attains 76.3% Precision@1 on post-2020 emerging skills. Dynamic-taxonomy metrics show higher coverage (0.942), relationship accuracy (0.876), and ∼3.2 × faster incorporation of new skills. Economic outcomes include a 42.8% reduction in mismatch, a 28.7% post-placement productivity gain, and a 20.9% increase in tenure; counterfactual simulations project 5.4–7.8 percentage-point reductions in structural unemployment and 6.8–10.5% productivity gains under realistic adoption.Originality/valueWe operationalize a continuously learning skill taxonomy embedded in matching policy via multi-agent reinforcement learning, bridging labor economics and AI. The approach is deployable within existing ATS/PES/EdTech workflows with governance. Limitations (uneven data coverage and interpretability) motivate integration with education systems and expanded fairness auditing.

Keywords: Deep reinforcement learning; Labor-market matching; Dynamic skill taxonomy; Multi-agent reinforcement learning; Technological transitions (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.emerald.com/ijm/article/47/2/193/13299 ... et-matching-via-deep (text/html)
https://www.emerald.com/ijm/article-pdf/47/2/193/11088329/ijm-07-2025-0595en.pdf (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eme:ijmpps:ijm-07-2025-0595

DOI: 10.1108/IJM-07-2025-0595

Access Statistics for this article

International Journal of Manpower is currently edited by Professor Adrian Ziderman

More articles in International Journal of Manpower from Emerald Group Publishing Limited
Bibliographic data for series maintained by Emerald Support ().