Akademik Veri Yönetim Sistemi
1st International Nature & Environment Conservation and Protected Areas Congress (INECPAC 2025), Antalya, Türkiye, 29 - 31 Ekim 2025, ss.513-518, (Tam Metin Bildiri)
Large language models (LLMs) have paved the way for making AI related tools accessible to ordinary users. A vast number of users benefit from LLMs using generative pre trained transformers (GPTs). GPTs help create content, translate languages, integrate agents into daily tasks, and handle specific queries. GPTs usually operate in a chatbot format and enable users to interact with them easily without requiring advanced computer skills. They can process text, voice, and image data. Currently, millions of people interact with chatbots to meet their specific needs, and the tourism industry has recently begun to emphasize their capabilities. Individual businesses and corporate companies use AI powered chatbots to answer user inquiries, such as product comparison, price checking, route planning, or providing onsite feedback. However, there is a need for refinement in these chatbots. (1) Chatbots rely on pretrained datasets; therefore, their information may not always be up to date. (2) Chatbots are trained for generalpurpose tasks, which means that they cannot perform well in domain specific queries such as cultural heritage, food and beverages, or history of places. This study adopts a retrieval augmented generation (RAG) approach to open source chatbots, which have great potential for revising and increasing their information capacity. RAG implementation highlights that ordinary chatbots struggle to answer questions regarding the St. Nicholaos and St. Nicholaos Church in Demre. Accordingly, a well trained chatbot can answer specific questions more accurately and reduce the occurrence of hallucinations.