Can Large Language Models Trigger a Paradigm Shift in Travel Behavior Modeling? Experiences with Modeling Travel Satisfaction

As a specific domain of subjective well-being, travel satisfaction has attracted much research attention recently. Previous studies primarily use statistical models and, more recently, machine learning models to explore the determinants of travel satisfaction. Both approaches require data from sufficient sample sizes and correct prior statistical assumptions. The emergence of Large Language Models (LLMs) offers a new modeling approach that can overcome the shortcomings of the existing methods. Pre-trained on extensive datasets, LLMs have strong capabilities in contextual understanding and generalization, significantly reducing their dependence on large quantities of task-specific data and stringent statistical assumptions. The primary challenge in applying LLMs lies in addressing the behavioral misalignment between LLMs and human behavior. Using data on travel satisfaction from a household survey in shanghai, this study identifies the existence and source of misalignment and develop methods to address the misalignment issue. We find that the zero-shot LLM exhibits behavioral misalignment, resulting in relatively low prediction accuracy. However, few-shot learning, even with a limited number of samples, allows the model to outperform baseline models in MSE and MAPE metrics. This misalignment can be attributed to the gap between the general knowledge embedded in LLMs and the specific, unique characteristics of the dataset. On these bases, we propose an LLM-based modeling approach that can be applied to model travel behavior using samples of small sizes. This study highlights the potential of LLMs for modeling not only travel satisfaction but also broader aspects of travel behavior.