Scholarlink

<p>Terrorist network analysis faces fundamental challenges in simultaneously capturing temporal evolution and multi-relational dependencies. Existing methods often treat operational, financial, and ideological relationships independently, overlooking their dynamic interdependencies. This study presents Terrorism Relation-aware Graph Neural Network, a novel framework integrating three core innovations: (1) a multi-scale temporal encoding scheme that distinguishes tactical, operational, and strategic timescales via a novel time-decay function; (2) a cross-relation attention mechanism that explicitly models influences among relationship types using learnable attention weights; and (3) a multi-relational graph construction that captures heterogeneous terrorist network structures. Comprehensive evaluation using time-stratified five-fold cross-validation on the Global Terrorism Database comprising over 200,000 incidents from 1970 to 2020 demonstrates statistically significant improvements (<i>p</i> &lt; 0.05, paired t-tests) over baseline models, including a 27.2% increase in attack type prediction accuracy, an AUC-ROC of 0.85 in organizational relationship mapping, and an F1-score of 0.79 in regional pattern detection. Ablation studies attribute 12.3% and 8.7% of performance gains to the cross-relation attention and temporal encoding components, respectively. TR-GNN maintains 71% prediction accuracy at year-long forecasting horizons, outperforming temporal baselines by over 15% points. Interpretability analyses reveal meaningful attention patterns that expose critical relationship-driven vulnerabilities. The framework also reduces false positive alerts by 38% and improves resource allocation efficiency by 45% during high-risk periods. These advances support data-driven threat anticipation, strategic planning, and evidence-based decision-making in counterterrorism efforts.</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

Graph neural networks for temporal terrorist network analysis and multi-relational threat prediction

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • Gideon Mwendwa,
  • Arpita Nagpal,
  • Sonia Garg,
  • Lokesh Chouhan,
  • Ngaira Mandela

摘要

Terrorist network analysis faces fundamental challenges in simultaneously capturing temporal evolution and multi-relational dependencies. Existing methods often treat operational, financial, and ideological relationships independently, overlooking their dynamic interdependencies. This study presents Terrorism Relation-aware Graph Neural Network, a novel framework integrating three core innovations: (1) a multi-scale temporal encoding scheme that distinguishes tactical, operational, and strategic timescales via a novel time-decay function; (2) a cross-relation attention mechanism that explicitly models influences among relationship types using learnable attention weights; and (3) a multi-relational graph construction that captures heterogeneous terrorist network structures. Comprehensive evaluation using time-stratified five-fold cross-validation on the Global Terrorism Database comprising over 200,000 incidents from 1970 to 2020 demonstrates statistically significant improvements (p < 0.05, paired t-tests) over baseline models, including a 27.2% increase in attack type prediction accuracy, an AUC-ROC of 0.85 in organizational relationship mapping, and an F1-score of 0.79 in regional pattern detection. Ablation studies attribute 12.3% and 8.7% of performance gains to the cross-relation attention and temporal encoding components, respectively. TR-GNN maintains 71% prediction accuracy at year-long forecasting horizons, outperforming temporal baselines by over 15% points. Interpretability analyses reveal meaningful attention patterns that expose critical relationship-driven vulnerabilities. The framework also reduces false positive alerts by 38% and improves resource allocation efficiency by 45% during high-risk periods. These advances support data-driven threat anticipation, strategic planning, and evidence-based decision-making in counterterrorism efforts.