EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

TalentFlow: Explainable Internal Talent Mobility Prediction Using Graph Representation Learning

Daniel Brooks, Priya Raman, Lina Chen

Academic Journal of Applied Sciences · 20.07.2026 · 2 (2) · 111-117

DOI: 10.54097/7zspnn42 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Academic Journal of Applied Sciences, online erschienen am 20.07.2026. Verfasst von Daniel Brooks und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Internal talent mobility—the movement of employees into new roles within the same organization—is central to workforce planning, yet most data-driven talent models treat each employee as an isolated feature vector and predict a single binary outcome such as promotion or attrition. This ignores a defining property of mobility: an employee’s next role is shaped not only by individual attributes but by relational context—the skills shared with target roles, the career ladders typical of a function, and the social capital embedded in a managerial and peer network. We present TalentFlow, a framework that reformulates internal mobility as next-role link prediction over a heterogeneous talent graph whose nodes are employees, roles, skills, departments and managers, connected by typed relations. A relational graph convolutional encoder learns employee and role representations by message passing over these relations, and a mobility decoder ranks candidate next roles for each employee. To make recommendations auditable, TalentFlow couples two complementary explanation modalities: SHAP attribution over employee attributes and a structural occlusion method that quantifies how much each relational channel contributed to a recommendation—an explanation flat-feature models cannot provide. Because real internal-mobility trajectories are proprietary, we construct TalentFlow-Sim, a benchmark that anchors employee attribute distributions to the public IBM HR Analytics dataset while generating a mobility layer driven by skill fit, career-ladder priors and managerial sponsorship. Across five seeds, relational message passing improves next-role ranking over strong non-graph baselines and over homogeneous-graph and structure-only embedding methods, and ablations show that removing the managerial/peer relation causes the largest degradation—direct evidence that social-capital signals carry predictive value recoverable only through the graph. TalentFlow complements recent explainable, fairness-aware promotion models such as FairPromote by extending explainable talent analytics from flat classification to relational, structurally grounded mobility prediction.

Zitieren

APA

Brooks, D., Raman, P., & Chen, L. (2026). TalentFlow: Explainable Internal Talent Mobility Prediction Using Graph Representation Learning. Academic Journal of Applied Sciences, 2(2), 111-117. https://doi.org/10.54097/7zspnn42

Vancouver

Brooks D, Raman P, Chen L. TalentFlow: Explainable Internal Talent Mobility Prediction Using Graph Representation Learning. Academic Journal of Applied Sciences. 2026;2(2):111-117. doi:10.54097/7zspnn42

BibTeX
@article{Brooks2026,
  author = {Brooks, Daniel and Raman, Priya and Chen, Lina},
  title = {TalentFlow: Explainable Internal Talent Mobility Prediction Using Graph Representation Learning},
  journal = {Academic Journal of Applied Sciences},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {111--117},
  doi = {10.54097/7zspnn42}
}
Kostenlos starten →