From narrative to action: A hierarchical cognitive large language model agent framework for interpretable and adaptive human mobility behavior generation

Understanding and replicating human mobility requires not only spatial–temporal accuracy but also an awareness of the cognitive hierarchy underlying real-world travel decisions. Existing approaches often lack interpretability and struggle to represent adaptive behavior under changing constraints. This study proposes a Narrative-to-Action hierarchical large language model agent framework that models human mobility as a cognitively interpretable, multi-level process. The framework decomposes mobility generation into macro-level narrative reasoning, meso-level reflective planning, and micro-level behavioral execution, enabling agents to produce semantically coherent and structurally executable daily schedules while allowing dynamic adjustment in response to contextual changes. A key component is a reflective mechanism drawing on Mobility Entropy by Occupation, which captures heterogeneous behavioral flexibility across occupational groups. Empirical evaluation using travel diary data from Guangzhou demonstrates improved fidelity across spatial, temporal, and semantic dimensions compared to baseline approaches. Overall, by explicitly modeling how individuals reason about and adapt to constraints, the framework provides a behaviorally grounded and interpretable approach for mobility analysis, providing a basis for more transparent future evaluation of policy interventions and equity impacts.