Scholarlink

<p>Accurate identification and timely correction of legal literacy misconceptions are essential for promoting civic competence and reducing behavioral risks among adolescents. However, existing educational data-mining approaches mainly rely on static statistical correlations, overlooking the dynamic, heterogeneous, and causally entangled nature of misconception formation. These limitations result in suboptimal intervention decisions and weak generalization across school environments. To overcome these challenges, this study proposes CGTCA-Net, a unified framework that integrates counterfactual causal reasoning with a spatiotemporal Graph-Transformer architecture to achieve dynamic tracking and adaptive optimization of legal literacy misconceptions. The framework first constructs a dynamic heterogeneous knowledge graph from multi-source school data and employs a Graph-Transformer backbone to capture temporal dependencies, semantic interactions, and cross-modal contextual cues. A counterfactual causal pruning mechanism grounded in the potential outcome framework is then designed to remove confounded edges and estimate individualized treatment effects with higher robustness. Finally, a multi-objective adaptive intervention module based on Pareto optimization generates personalized, cost-effective, and fairness-aware educational recommendations. Extensive experiments conducted across multiple school scenarios demonstrate that CGTCA-Net outperforms state-of-the-art baselines in entity recognition (+ 7.3%), causal effect estimation accuracy (+ 11.2%), misconception trajectory prediction (+ 9.6%), and intervention effectiveness (+ 13.4%). Ablation studies further validate the contribution of the causal pruning strategy and the cross-modal semantic fusion module. The results confirm that CGTCA-Net provides a reliable and interpretable solution for precision legal education, offering practical value for real-world deployment in intelligent tutoring, campus governance, and rule-of-law curriculum design.</p> Graphical Abstract <p></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

A counterfactual graph transformer framework for dynamic tracking and adaptive optimization of legal literacy misconception interventions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月3日

查看全文
  • Juan Wang,
  • Yongguan Ai

摘要

Accurate identification and timely correction of legal literacy misconceptions are essential for promoting civic competence and reducing behavioral risks among adolescents. However, existing educational data-mining approaches mainly rely on static statistical correlations, overlooking the dynamic, heterogeneous, and causally entangled nature of misconception formation. These limitations result in suboptimal intervention decisions and weak generalization across school environments. To overcome these challenges, this study proposes CGTCA-Net, a unified framework that integrates counterfactual causal reasoning with a spatiotemporal Graph-Transformer architecture to achieve dynamic tracking and adaptive optimization of legal literacy misconceptions. The framework first constructs a dynamic heterogeneous knowledge graph from multi-source school data and employs a Graph-Transformer backbone to capture temporal dependencies, semantic interactions, and cross-modal contextual cues. A counterfactual causal pruning mechanism grounded in the potential outcome framework is then designed to remove confounded edges and estimate individualized treatment effects with higher robustness. Finally, a multi-objective adaptive intervention module based on Pareto optimization generates personalized, cost-effective, and fairness-aware educational recommendations. Extensive experiments conducted across multiple school scenarios demonstrate that CGTCA-Net outperforms state-of-the-art baselines in entity recognition (+ 7.3%), causal effect estimation accuracy (+ 11.2%), misconception trajectory prediction (+ 9.6%), and intervention effectiveness (+ 13.4%). Ablation studies further validate the contribution of the causal pruning strategy and the cross-modal semantic fusion module. The results confirm that CGTCA-Net provides a reliable and interpretable solution for precision legal education, offering practical value for real-world deployment in intelligent tutoring, campus governance, and rule-of-law curriculum design.

Graphical Abstract