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

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.

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

The paper "Where You Go is Who You Are: Behavioral Theory-Guided LLMs for Inverse Reinforcement Learning" introduces the SILIC framework, which utilizes LLMs in conjunction with Inverse Reinforcement Learning (IRL) and Cognitive Chain Reasoning (CCR) to infer sociodemographic attributes from human mobility patterns. This research addresses the inherent limitations of existing trajectory datasets, which often lack critical traveler sociodemographic information, thereby constraining their utility in contexts such as transportation planning and policy formulation.

The authors explicitly integrate the Theory of Planned Behavior (TPB) to model the cognitive processes behind travel decision-making, thereby capturing latent behavioral intentions and sociodemographic attributes. SILIC leverages LLMs to enhance the IRL process by providing heuristics for reward function initialization and updates, addressing the traditional challenges associated with IRL, including its ill-posed nature and optimization difficulties in a vast and unstructured reward space.

Methodology

The study is structured around two main components:

1. LLM-guided Inverse Reinforcement Learning (IRL):
   • The IRL framework is designed to infer individual-specific reward functions that encapsulate underlying behavioral intentions from observed trajectories.
   • LLMs are utilized for the initialization and iterative updating of reward weights, which is critical to overcoming IRL's ill-posed challenges and ensuring convergence to plausible behavioral solutions.
2. Cognitive Chain Reasoning (CCR):
   • CCR applies the TPB in reverse to predict sociodemographic attributes. After inferring latent intentions using IRL, CCR subsequently aligns these with sociodemographic predictors by reasoning through belief constructs influenced by environmental contexts.

Empirical Evaluation

The SILIC framework was evaluated using data from the 2017 Puget Sound Regional Council Household Travel Survey. The results indicated that SILIC significantly outperformed state-of-the-art baseline models in predicting various sociodemographic attributes such as gender, age, and employment status.

For example, in gender prediction tasks, SILIC improved accuracy by over 30% compared to benchmarks, highlighting the efficacy of integrating psychological constructs via LLMs into IRL processes. Similarly, SILIC excelled in age prediction, effectively identifying patterns where traditional methods faced challenges due to data imbalance or oversimplification.

Implications and Future Directions

The proposed SILIC framework expands the scope of trajectory data analysis by enriching them with inferred sociodemographic insights, thus aiding in more informed transportation planning, targeted marketing, and crisis management applications. The integration of cognitive modeling with LLMs sets a precedent for more nuanced behavioral analytics, potentially transforming how mobility datasets are utilized.

Future research could explore refining LLM heuristics for specific geographic or demographic trends and enhancing the fidelity of inferred cognitive constructs. Additionally, extending the framework to accommodate more nuanced state-space representations could further improve the model's capacity to mirror real-world decision-making processes.

In conclusion, this research presents a compelling blend of LLM capabilities with foundational behavioral theories to advance the field of human mobility analysis, offering a robust methodological contribution to both the transportation and AI communities.