Scholarlink

<p>Green supply chains increasingly rely on ESG disclosures, regulatory texts and public narratives, yet most existing risk-analytics approaches focus on structured indicators and struggle to turn noisy ESG text into interpretable risk signals. This study proposes the Textual Analytics and Predictive Risk Assessment Model (TAPRAM), an LLM-enabled framework that integrates ESG-related textual streams into probabilistic risk assessment for green supply chains. TAPRAM uses a large language model encoder to derive contextual ESG representations, a multi-task prediction head to jointly classify risk facet and severity (Low/Medium/High), and an editable semantic-rule layer that allows domain experts to refine or override model outputs in line with evolving regulations. To evaluate the framework under realistic but confidential conditions, we generate a statistically validated synthetic risk stream via a Gaussian-copula with AR(1) temporal structure that is distributionally indistinguishable from the EcoTrans-2023 transaction log. In theoretical terms, TAPRAM positions ESG-text–driven, LLM-enabled risk identification as a coherent framework for interpretable green-supply-chain analytics that links textual ESG signals to probabilistic risk assessments. In practical terms, it provides a deployable decision-support tool that helps firms prioritise audits and green investments under tightening sustainability regulations, operationalising ESG-aware dashboards, audit-prioritisation grids and regulatory “what-if” sandboxes in day-to-day risk governance. Experiments on the synthetic stream show that TAPRAM achieves an accuracy of 92.3%, macro-F1 of about 0.91 and AUC of about 0.95, significantly outperforming strong machine-learning and LLM baselines, indicating that ESG-text-driven analytics can materially enhance green supply-chain risk management.</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

Leveraging large language models for enhanced risk management in green supply chains

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Yan Liu,
  • Susheng Wang,
  • Peng Zhang

摘要

Green supply chains increasingly rely on ESG disclosures, regulatory texts and public narratives, yet most existing risk-analytics approaches focus on structured indicators and struggle to turn noisy ESG text into interpretable risk signals. This study proposes the Textual Analytics and Predictive Risk Assessment Model (TAPRAM), an LLM-enabled framework that integrates ESG-related textual streams into probabilistic risk assessment for green supply chains. TAPRAM uses a large language model encoder to derive contextual ESG representations, a multi-task prediction head to jointly classify risk facet and severity (Low/Medium/High), and an editable semantic-rule layer that allows domain experts to refine or override model outputs in line with evolving regulations. To evaluate the framework under realistic but confidential conditions, we generate a statistically validated synthetic risk stream via a Gaussian-copula with AR(1) temporal structure that is distributionally indistinguishable from the EcoTrans-2023 transaction log. In theoretical terms, TAPRAM positions ESG-text–driven, LLM-enabled risk identification as a coherent framework for interpretable green-supply-chain analytics that links textual ESG signals to probabilistic risk assessments. In practical terms, it provides a deployable decision-support tool that helps firms prioritise audits and green investments under tightening sustainability regulations, operationalising ESG-aware dashboards, audit-prioritisation grids and regulatory “what-if” sandboxes in day-to-day risk governance. Experiments on the synthetic stream show that TAPRAM achieves an accuracy of 92.3%, macro-F1 of about 0.91 and AUC of about 0.95, significantly outperforming strong machine-learning and LLM baselines, indicating that ESG-text-driven analytics can materially enhance green supply-chain risk management.