Paper Title
Establishing Intelligent Knowledge-Based Dialogue Techniques for Environmental Decision Management

Abstract
Generative AI can help users efficiently extract relevant information from large volumes of textual data and organize it into comprehensible statements, thereby improving the effectiveness of knowledge sharing and application. Transforming data and information into valuable knowledge often requires processes of learning and value recognition. This study aims to assist the local government in leveraging its extensive collection of research reports, regulations, case studies, and technical manuals by developing a systematic approach to collecting, organizing, managing, sharing, and applying this information. Ultimately, the goal is to facilitate internal knowledge sharing and support management and decision-making processes. Environmental issues often stem from a variety of interconnected events and cases. By learning from past experiences, it is possible to prevent future soil and groundwater contamination incidents and address new cases more efficiently. As part of this study, we focused on research topics related to soil and groundwater management and developed a soil and groundwater knowledge dialogue technology. It incorporates keyword extraction, knowledge association networks, topic-based knowledge ontology networks, and large language models (LLMs). The results show that topic-based knowledge ontology networks effectively enhance the accuracy of LLM-based question-answering, reducing instances of irrelevant responses. In LLMs training, the loss decreases to 0.2, but loss of LLMs testing is increased from 2.2 to 2.6. Although, the technique provides corresponding reference materials, the answer is not reliability. This study integrated Retrieval Augmented Generation to our model to increase accuracy and expertise. Our study technique provides corresponding reference materials, enabling users to verify the reliability of the answers.To evaluate the technique, government personnel were invited to participate in testing. The results demonstrated that the technique consistently provided accurate answers aligned with the content of the reports, with no instances of irrelevant responses. This indicates that the technique can serve as a valuable tool for training and educating personnel or for scientifically informed strategy development within relevant departments. Keywords - Generative Artificial Intelligence, Fine-Tuning, Chatbots, Ontology