Where You Go is Who You Are: Behavioral Theory-Guided LLMs for Inverse Reinforcement Learning

TL;DR

This work tackles inferring sociodemographic attributes from mobility trajectories lacking metadata, addressing the need to uncover latent cognitive drivers behind travel behavior. It introduces SILIC, a two-stage framework that first uses LLM-guided inverse reinforcement learning to recover individualized intentions under the Theory of Planned Behavior, then applies Cognitive Chain Reasoning to map beliefs to attributes. The method leverages contextual variables and LLM priors to initialize and refine reward functions, mitigating IRL's ill-posedness and improving interpretability. Evaluated on the 2017 Puget Sound GPS dataset, SILIC significantly outperformed baselines in predicting gender and age, demonstrating potential for enriching big trajectory data for transportation planning and behaviorally grounded AI systems.

Abstract

Big trajectory data hold great promise for human mobility analysis, but their utility is often constrained by the absence of critical traveler attributes, particularly sociodemographic information. While prior studies have explored predicting such attributes from mobility patterns, they often overlooked underlying cognitive mechanisms and exhibited low predictive accuracy. This study introduces SILIC, short for Sociodemographic Inference with LLM-guided Inverse Reinforcement Learning (IRL) and Cognitive Chain Reasoning (CCR), a theoretically grounded framework that leverages LLMs to infer sociodemographic attributes from observed mobility patterns by capturing latent behavioral intentions and reasoning through psychological constructs. Particularly, our approach explicitly follows the Theory of Planned Behavior (TPB), a foundational behavioral framework in transportation research, to model individuals' latent cognitive processes underlying travel decision-making. The LLMs further provide heuristic guidance to improve IRL reward function initialization and update by addressing its ill-posedness and optimization challenges arising from the vast and unstructured reward space. Evaluated in the 2017 Puget Sound Regional Council Household Travel Survey, our method substantially outperforms state-of-the-art baselines and shows great promise for enriching big trajectory data to support more behaviorally grounded applications in transportation planning and beyond.

Figures (7)

• Figure 1: Overview of our proposed framework. We inversely follow the Theory of Planned Behavior (TPB) to predict sociodemographic attributes from travel trajectories. Our method first uses an LLM-guided IRL model to infer behavioral intentions, followed by a Cognitive Chain Reasoning strategy that predicts sociodemographic attributes via intermediate belief constructs.
• Figure 2: Methodological Framework.
• Figure 3: F1 score variation across feature selection thresholds for gender prediction using SVM, XGBoost, and CatBoost.
• Figure 4: F1 score variation across feature selection thresholds for age prediction using SVM, XGBoost, and CatBoost.
• Figure 5: Prompt used to initialize reward weights for IRL