Reading between the lines: Using large language models to uncover causal pathways in transport policy

Understanding the relational mechanisms behind transport policy outcomes is essential for informed policy design and evaluation. This paper introduces a novel framework that harnesses large language models (LLMs) to extract, construct, and analyze directed association networks from unstructured news text. By identifying policy-related events and their interlinked consequences, the framework enables the systematic mapping of policy association pathways within transportation governance. Using a large-scale corpus of over 8 million news articles, we demonstrate how LLMs can transform raw textual data into structured relational representations, which are subsequently analyzed to uncover hidden patterns, interdependencies, and the real-world impacts of policy interventions. In our case study, the constructed transport policy relational network demonstrates properties commonly observed in complex social systems. Specifically, the network exhibits a scale-free topology, evidenced by a heavy-tailed total degree distribution and the emergence of high-degree hub nodes that serve as central connectors within the system, while in- and out-degree distributions remain heterogeneity. Furthermore, the study reveals a well-defined modular structure, indicating the presence of tightly interconnected clusters of policy-relevant variables or events. Building on this structural analysis, we further identify a set of influential variables and association pathways that play pivotal roles in shaping transport policy dynamics. The proposed approach offers a scalable and interpretable method for identifying policy-driven factors and tracing their downstream effects, highlighting the potential of generative AI to support evidence-based policy analysis and explainable policy design.