Scholarlink

<p>Digital skills are crucial for career adaptability and long-term labour market outcomes, yet identifying interventions that deliver sustained benefits while ensuring equity remains challenging. To address this, we propose CausalSkillNet-RL, a fairness-aware causal reinforcement learning framework for personalised intervention recommendations. The framework combines for the first time (i) a novel fairness-aware reinforcement learning agent that optimises both immediate uplift and long-term skill growth while enforcing equity for protected groups; and (ii) a fairness-regularised DragonNet for principled estimation of heterogeneous treatment effects, to provide robust causal signals for policy learning. We evaluate CausalSkillNet-RL firstly on synthetic and real-world datasets (Stack OverFlow Survey, ClickStart), using heterogeneous effect metrics, ablation studies, and visualisation. Results show that it achieves higher average treatment effects, improved Uplift@K, and reduced fairness gaps compared to state-of-the-art baselines. These findings highlight the framework’s potential to support equitable, data-driven long-term interventions in digital skill development.</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

Towards fair evaluation of digital skills training on career adaptability and outcomes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Zhifei Hu,
  • Alexandra I. Cristea,
  • Sue Black

摘要

Digital skills are crucial for career adaptability and long-term labour market outcomes, yet identifying interventions that deliver sustained benefits while ensuring equity remains challenging. To address this, we propose CausalSkillNet-RL, a fairness-aware causal reinforcement learning framework for personalised intervention recommendations. The framework combines for the first time (i) a novel fairness-aware reinforcement learning agent that optimises both immediate uplift and long-term skill growth while enforcing equity for protected groups; and (ii) a fairness-regularised DragonNet for principled estimation of heterogeneous treatment effects, to provide robust causal signals for policy learning. We evaluate CausalSkillNet-RL firstly on synthetic and real-world datasets (Stack OverFlow Survey, ClickStart), using heterogeneous effect metrics, ablation studies, and visualisation. Results show that it achieves higher average treatment effects, improved Uplift@K, and reduced fairness gaps compared to state-of-the-art baselines. These findings highlight the framework’s potential to support equitable, data-driven long-term interventions in digital skill development.