Can Large Language Models Understand Dynamic Joint Decision-Making Processes? Evidence on Destination Choice for Leisure Activities

Abstract

This study examines the potential of large language models (LLMs) to understand joint travel decisions related to social activities, using group chat data from messaging platforms like WhatsApp. These decisions involve dynamic negotiations, considering the preferences and constraints of each participant. To fully understand the decision-making process, it is necessary to infer nuanced and implicit information from the social and cultural context of each generation and country. Specifically, decision-making factors must be represented in a structured format, requiring extensive humanlabeled annotations. A customized prompt, based on chain-of-thought reasoning, is designed to first identify individual/clique characteristics and then trace how travel decisions evolve through iterative negotiations, capturing both explicit and implicit factors from Japanese and Korean data. We quantitatively evaluate the performance of LLMs in structuring the data, identify optimal prompting methods, and recognize situations where LLMs struggle. These findings highlight the potential of LLM-based analysis for enhancing context-rich activity-based modeling.