Large Language Model-Driven Framework for Dynamic Stakeholder-Associated Issues Analysis in Mega Infrastructure Projects
本研究提出一种结合大型语言模型和主题随时间变化模型的动态分析方法，自动识别利益相关者及其与相关问题的互动，并通过香港机场三跑道系统14年案例验证，为重大基础设施项目提供分阶段、针对性的利益相关者参与策略。
The complexity and dynamic nature of megaprojects necessitate effective stakeholder engagement strategies to ensure the successful achievement of project objectives. However, the lack of an effective framework to uncover interactions and trends between stakeholders and associated issues within the vast, encapsulated project data is a key factor hindering dynamic stakeholder engagement. Traditional methods are constrained by time-consuming, limited semantic interpretability, accuracy issues, and weak reasoning capabilities, making them insufficient for addressing dynamic stakeholder engagement challenges. To address this gap, this research introduced an integrated LLM-TOT (Large Language Model-Topic over Time) dynamic analysis method that enables the automated identification of stakeholders and the analysis of their interactions with relevant issues over time, prioritizing stakeholders directly linked to relevant issues at each project phase. Dynamic Stakeholder Engagement Strategies Framework was proposed to provide efficient and adaptive stakeholder engagement strategies for mega infrastructure projects (MIPs). Using a 14-year case study of the Hong Kong International Airport Third Runway System (HKIA 3RS) project, this study validated the proposed framework by analyzing 1,323 official project documents. The analysis identified 16 critical stakeholder-associated issues, 43 key stakeholders, and their relevance, which were systematically scored. Two validation experiments were conducted to rigorously evaluate the accuracy and reliability of the results derived from the LLM-TOT dynamic analysis method. Based on these findings, a stakeholder engagement framework tailored to the HKIA 3RS was developed and verified against real-world practices, offering strategies for phase-specific, issue-level, and dynamic engagement. This study introduced an integrated LLM-TOT dynamic analysis method as an innovative application that connects a standard TOT model with a commercially available LLM to analyze stakeholder-issue dynamics in MIPs. In addition, it presented a Dynamic Stakeholder Engagement Framework that equips decision-makers with targeted, systematic, and actionable strategies to address the “how-to” challenges of dynamic stakeholder engagement throughout the lifecycle of MIPs.