Presentation Information
[3Yin-A-60]A proposal-based tourist information system using Large Language Models
〇Ryosei Maeda1, Tetsuro Takahashi1 (1. Kagoshima University)
Keywords:
Large Language Models,RAG,Sightseeing
We developed an interactive travel agent specialized for tourism in a specific region. We generated tourist guide text using a Large Language Model (LLM) with a Retrieval-Augmentation Generation (RAG) that utilized 2,370 sights created by the Kagoshima Prefecture Tourism Federation. We also implemented a simulation LLM mechanism to verify the feasibility of the generated plans. This suppresses hallucination while presenting highly feasible travel plans that take travel time and length of stay into account.
Quantitative evaluation using 155 evaluation data points demonstrated that the tourist guide system implemented in this study demonstrated significant improvements in information accuracy and plan feasibility compared to an existing LLM, demonstrating a significant reduction in the presentation of nonexistent spots and infeasible itineraries. Furthermore, qualitative evaluation by subjects confirmed that the time required for plan creation and the psychological burden were reduced compared to traditional planning using internet search. This study demonstrates a method for building a next-generation tourism agent that combines reliability and practicality, providing useful insights for promoting digital transformation in regional tourism.
Quantitative evaluation using 155 evaluation data points demonstrated that the tourist guide system implemented in this study demonstrated significant improvements in information accuracy and plan feasibility compared to an existing LLM, demonstrating a significant reduction in the presentation of nonexistent spots and infeasible itineraries. Furthermore, qualitative evaluation by subjects confirmed that the time required for plan creation and the psychological burden were reduced compared to traditional planning using internet search. This study demonstrates a method for building a next-generation tourism agent that combines reliability and practicality, providing useful insights for promoting digital transformation in regional tourism.