GeoMinLM: A Large Language Model in Geology and Mineral Survey in Yunnan Province
Yu Fu, Mingguo Wang, Chengbin Wang, Shangxin Dong, Jianguo Chen, Jiyuan Wang, Hongping Yu, Jing Huang, Liheng Chang, Bo Wang
Abstract
• Build a LLM for Geology and Mineral Survey. • The knowledge infusion from knowledge graph for improving the performance of GeoMinLM. • GeoMinLM is useful to provide the knowledge for geology and mineral survey. In recent years, the development of artificial intelligence and big data technologies has led to the advancement of tools and solutions for transforming the geological and mineral survey paradigm, which requires a large amount of geological knowledge in a complex and arduous working environment. The large language model (LLM) has a significant advantage in answering generative intelligent questions. However, LLMs for general fields have limitations in answering professional questions in a vertical domain like geology. To overcome this challenge, we proposed and developed GeoMinLM, an LLM for geological and mineral exploration scenarios in Yunnan Province, and explored its applications in intelligent Q&A. Leveraging a proprietary dataset of 5.16 million words in geology and mineral exploration, we trained GeoMinLM based on Baichuan-2, achieving superior performance through fine-tuning and hyperparameter optimization. By integrating expert knowledge via a knowledge graph, we significantly reduced hallucinations and enhanced professionalism. This study proves that GeoMinLM is helpful for accurate information retrieval and knowledge dissemination, thereby supporting the intelligent advancement of geological and mineral fields.