ChatPlanner Uses LLMs for Personalized Public Transit Routing

Researchers have developed ChatPlanner, an innovative framework that utilizes large language models (LLMs) to create personalized public transit routing solutions. The core challenge addressed is the difficulty of accurately capturing and integrating diverse user preferences into traditional routing algorithms. ChatPlanner aims to overcome this by allowing users to express their preferences through natural language queries. The framework employs fine-tuned LLMs combined with Retrieval-Augmented Generation (RAG) to effectively extract necessary routing parameters and interpret the subtle nuances of user preferences from these natural language inputs. These interpreted preferences are then seamlessly integrated into the objective function of a public transit routing algorithm, enabling truly personalized route generation. To validate its capabilities, the study designed preference-aware datasets incorporating eight distinct personas and five contexts, establishing clear scoring standards for both fine-tuning and RAG. Experiments confirmed ChatPlanner's ability to reliably generate feasible solutions, with fine-tuning enforcing output structure and learning general preference patterns, while RAG provides query-specific context for imprecise expressions. This hybrid approach achieved the highest accuracy in extracting routing information and interpreting user preferences, demonstrating its potential to identify valuable route alternatives often overlooked by existing planners.