LLM-driven rapid construction of knowledge graph for mineral resources: a case study of the Dajishan hydrothermal tungsten deposit

The volume of mineral resources big data is rapidly increasing due to geological exploration and mining activities. Such data are characterized by multi-source heterogeneity, complex structures, and unclear interrelationships. Knowledge Graph (KG), with its powerful capabilities in knowledge structuring, semantic association, and intelligent reasoning, is being increasingly applied in the field of mineral resources, highlighting its growing importance. This paper focuses on methodological innovations for the rapid and efficient construction of KG in the mineral resources domain. Taking strategically significant hydrothermal tungsten deposits as the research subject, and addressing the limitations of traditional KG construction methods in terms of efficiency, automation, and processing massive unstructured text, we introduce large language model (LLM) technology to develop a rapid KG construction framework characterized by “LLM-driven approach guided by mineral resources knowledge.” A case study of the Dajishan Tungsten Mine KG was conducted, and the TOPSIS method was employed to deeply explore its prospecting indicators. By integrating direct prospecting indicators, potential prospecting indicators, and auxiliary prospecting information, a prospecting model for hydrothermal tungsten deposits in the Nanling region was established, providing valuable references for mineral exploration. Furthermore, a knowledge base constructed based on this KG demonstrates significant improvements in various capabilities, particularly in deep reasoning, compared to traditional knowledge bases. The research shows that the KG, developed through the integration of LLM technology, not only greatly enhances the speed and scale of integrating key information but also enables in-depth mining of prospecting information, demonstrating strong application potential.