Smart Tourism Information QA Service using the Prompt and Generative AI Foundation Model

다국어 초록 (Multilingual Abstract)

After COVID-19, travel patterns have shifted from group tours to individual tourists who plan their own trips.
Therefore, the development of generative AI-based, personalized smart tourism services for individual tourists is necessary. We are developing a smart tourism service platform that provides tourism information, travel planner services, and tour guide services using an AI-based chatbot service. In this paper, we develop the smart tourism QA chatbot service for individual tourists, using ChatGPT-4o-mini and RAG. The RAG system must accurately identify the intent of the user's question and generate high-quality prompts to mitigate the Hallucination and Sovereign AI problems inherent in generative AI. In this paper, we propose the RAG system that uses existing BERT-based NER and DST models, the tourism information MySQL DB, and the tourism information knowledge base implemented with a Neo4J graph DB to improve accuracy. The NER and DST models manage the user's conversation state, identify the intent of the question, and search the tourism information knowledge base to generate high-quality prompts. The performance of the proposed RAG system and ChatGPT-4o-mini is analyzed using the previously developed tourism information QA training dataset, and the results show an excellent performance of 99.54%. The sovereign AI problem can be mitigated by using the NER model that applies forbidden words to identify the user's question intent. The smart tourism information QA service proposed in this paper can be used to develop travel planner and tour guide services.