Scholarlink

<p>In the contemporary era of digital transformation, large enterprises are confronted with three significant challenges: the presence of isolated human resource data, the necessity of privacy compliance, and the imperative for dynamic decision-making. Conventional static models are ill-equipped to capture the evolution of skills and organisational changes, while regulations such as the General Data Protection Regulation (GDPR) impede data circulation. In order to address the aforementioned issues, we hereby introduce Fed Hyper KG-HR, a novel framework that pioneers the integration of hypergraph neural networks and federated learning. This innovative development solves, for the first time, these core HR challenges. A significant innovation is the four-dimensional spatiotemporal ontology model, which has been shown to capture complex cross-organizational relationships with greater accuracy, achieving 18.3% higher collaborative identification accuracy than traditional federated learning methods. The innovative temporal reasoning mechanism underpinning our dynamic decision-making process represents a significant advancement in the field of skill modelling. This breakthrough has been demonstrated to achieve a promotion prediction accuracy that exceeds 93%, while concurrently reducing false positive rates in resignation warnings to below 3%. This represents a substantial enhancement, with a 50% improvement over conventional industrial systems. In the field of privacy protection, our innovative adaptive perturbation technique has been demonstrated to achieve an accuracy of 89% in decision-making, while maintaining a zero data leakage (Adaptive perturbation technology achieves 89% decision accuracy at ε=1.2, with controllable privacy leakage risk) rate. This enhancement in efficiency, as measured by the accuracy-to-time ratio, has been shown to exceed 35% when compared to conventional baseline methods. Empirical evidence has demonstrated a growth in cross-departmental resource matching efficiency that exceeds 50%. This establishes a ‘knowledge-driven’ paradigm, which is unique, and exhibits excellent transferability across industries, such as finance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on the construction of dynamic knowledge graph and intelligent decision-making for enterprise human resource information based on federated hyper graph neural network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月27日

查看全文
  • Zhenmei Wu,
  • Hong Zhang

摘要

In the contemporary era of digital transformation, large enterprises are confronted with three significant challenges: the presence of isolated human resource data, the necessity of privacy compliance, and the imperative for dynamic decision-making. Conventional static models are ill-equipped to capture the evolution of skills and organisational changes, while regulations such as the General Data Protection Regulation (GDPR) impede data circulation. In order to address the aforementioned issues, we hereby introduce Fed Hyper KG-HR, a novel framework that pioneers the integration of hypergraph neural networks and federated learning. This innovative development solves, for the first time, these core HR challenges. A significant innovation is the four-dimensional spatiotemporal ontology model, which has been shown to capture complex cross-organizational relationships with greater accuracy, achieving 18.3% higher collaborative identification accuracy than traditional federated learning methods. The innovative temporal reasoning mechanism underpinning our dynamic decision-making process represents a significant advancement in the field of skill modelling. This breakthrough has been demonstrated to achieve a promotion prediction accuracy that exceeds 93%, while concurrently reducing false positive rates in resignation warnings to below 3%. This represents a substantial enhancement, with a 50% improvement over conventional industrial systems. In the field of privacy protection, our innovative adaptive perturbation technique has been demonstrated to achieve an accuracy of 89% in decision-making, while maintaining a zero data leakage (Adaptive perturbation technology achieves 89% decision accuracy at ε=1.2, with controllable privacy leakage risk) rate. This enhancement in efficiency, as measured by the accuracy-to-time ratio, has been shown to exceed 35% when compared to conventional baseline methods. Empirical evidence has demonstrated a growth in cross-departmental resource matching efficiency that exceeds 50%. This establishes a ‘knowledge-driven’ paradigm, which is unique, and exhibits excellent transferability across industries, such as finance.