Geological Knowledge Extraction and Information Mining via the Fusion of Knowledge Graphs and LLMs: A Case Study of Carlin-Type Gold Deposits
-
摘要:
针对地质勘查领域海量非结构化数据难以被有效利用以及通用大模型存在“事实幻觉”与专业逻辑匮乏等问题，本文提出了一种融合知识图谱（KG）与检索增强生成（RAG）的垂直领域智能知识挖掘框架，并以中国黔西南与美国内华达地区的卡林型金矿成矿规律总结和对比研究为例进行了验证.首先，构建了基于本地化部署DeepSeek-32B的RAG智能问答系统，通过向量检索与生成式阅读理解，实现了专业知识的精准溯源与高可信问答.其次，利用大模型监督微调（SFT）技术，从数百份多源异构地质资料中高效构建了系统涵盖地层构造、蚀变矿物及控矿要素的跨区域成矿知识图谱.实验结果表明，该系统在客观准确性上显著优于GPT-4o，在主观生成上具备高忠实度与完全可溯源性，有效解决了幻觉问题.基于图谱拓扑学的分析不仅定量揭示了两地成矿的宏观异同，还量化了从矿石实体、蚀变组合到地球化学元素异常的级联指示路径，证实了其发现隐性找矿线索的能力.该研究实现了从非结构化文本到结构化知识的智能转化与深度挖掘，为解决地学领域“海量数据、知识饥饿”困境提供了新的技术路径.
Abstract:To address the challenges in effectively utilizing massive unstructured data within geological exploration and the issues of hallucination and lack of specialized logic in general Large Language Models (LLMs), we propose an intelligent knowledge mining framework for vertical domains by integrating Knowledge Graph (KG) and Retrieval-Augmented Generation (RAG). This framework is validated through a comparison case study of Carlin-type gold deposits in the Southwest Guizhou, China, and in Nevada, USA. Firstly, a RAG-based intelligent question-answering system was constructed using a locally deployed DeepSeek-32B model. Through vector retrieval and generative reading comprehension, the system achieved precise traceability of professional knowledge and highly reliable Question & Answer (Q&A). Secondly, leveraging Supervised Fine-Tuning (SFT) techniques on the LLM, we developed a cross-regional metallogenic knowledge graph systematically covering stratigraphy, structure, alteration minerals, and ore-controlling factors based on hundreds of multi-sources, and heterogeneous geological documents. The results demonstrate that the proposed system significantly outperforms GPT-4o in terms of objective accuracy. For subjective content generation, it exhibits high faithfulness, full traceability and effectively mitigate the hallucination. Analyses based on graph topology not only quantitatively reveal the macroscopic similarities and differences of Au mineralization between the two regions but also quantify the cascading indicative pathways-from orebody entities and alteration assemblages to geochemical element anomalies, confirming the system's capability to discover implicit clues for mineral exploration. This study realizes the intelligent transformation and in-depth mining of knowledge from unstructured text to structured representations, offering a novel technical pathway to address the dilemma of "data-rich yet knowledge-poor" prevalent in the geoscience domain.
-
图 1 美国内华达地区（a）与中国右江盆地（b）卡林型金矿床分布(据Wang et al., 2018修改）
Fig. 1. Distribution of Carlin-type gold deposits in (a) Nevada of the USA and (b) the Youjiang basin of China (modified from Wang et al., 2018)
图 2 大语言模型与知识图谱双驱动的知识抽取与信息挖掘架构
Fig. 2. A dual-driven architecture for knowledge extraction and information mining based on LLMs and knowledge graphs
图 3 中国黔西南‒美国内华达卡林型金矿知识图谱本体层架构
图示中为视觉简洁，将连接不同属性实体（如规模、品位、特征等）的多种具体关系类型统一展示为通用标签“具有”
Fig. 3. Ontology layer architecture of knowledge graph for Carlin-type gold deposits in SW Guizhou of China and Nevada of the USA
图 4 知识抽取模型微调前后性能对比
Fig. 4. Performance comparison of the knowledge extraction model before and after fine-tuning
图 6 中国黔西南‒美国内华达卡林型金矿知识图谱（部分展示）
Fig. 6. Knowledge graph of Carlin-type gold deposits in Southwest Guizhou of China and Nevada of the USA (partial)
图 7 客观题回答效果评估数据集与示例试题
Fig. 7. Dataset and sample items for performance evaluation of objective question answering
图 8 基于知识图谱的矿床社区发现聚类分析结果
Fig. 8. Cluster analysis results of ore deposit community detection based on KGs
图 10 黔西南典型矿床‒围岩蚀变‒地球化学异常指示关系
Fig. 10. Indicative relationships between typical deposits, wall-rock alterations, and geochemical anomalies in Southwest Guizhou
表 1 中国黔西南‒美国内华达卡林型金矿知识图谱实体关系统计
Table 1. Statistics of entity relations in the knowledge graph for Carlin-type gold deposits in SW Guizhou of China and Nevada of the USA
区域 实体类型 实体个数 实体个数（合并消歧后） 关系类型 关系个数（合并消歧后） 中国黔西南 9 10 212 5 234 12 10 256 美国内华达 8 2 312 987 11 1 523 合计 9 12 524 6 221 12 11 779 
下载: 导出CSV
表 2 垂直领域大模型客观题评估
Table 2. Objective evaluation of domain-specific LLMs
模型名称 Part1 Part2 Part3 Part4 Part5 总得分 Gemini2.5-Flash 0.85 0.9 1 1 1 0.95 Gemini2.5-Pro 0.95 0.95 1 0.95 1 0.97 Qwen3-Max 0.92 0.95 0.93 1 1 0.95 Qwen2.5 0.65 0.75 0.84 0.95 0.96 0.81 DeepSeek-R1 0.85 0.95 0.92 1 1 0.94 DeepSeek-V3 0.80 0.80 1 1 0.95 0.91 豆包-1.6 0.91 0.95 1 1 1 0.97 GPT4.1 mini 0.85 0.95 0.90 1 1 0.94 GPT4o 0.91 0.80 1 1 1 0.94 GeoGPT-R1-Preview 0.85 0.95 1 1 0.90 0.94 DeepSeek-32B-Base 0.85 0.91 0.85 0.92 0.90 0.88 RAG-黔西南 0.95 1 0.98 0.99 1 0.99
下载: 导出CSV
表 3 垂直领域大模型主观性能评估
Table 3. Subjective performance evaluation of domain-specific LLMs
模型名称 忠实度 答案相关性 上下文精确度 知识溯源性 综合评价 GPT-4o 0.72 0.92 - 0.00 语言组织流畅，但在处理特定矿区数据时存在明显事实幻觉，且无法溯源. GeoGPT-R1-Preview 0.78 0.88 0.65 0.40 具备地学常识与推理能力，但因缺乏本地私有文献支持，细节准确度不足. DeepSeek-R1-32B-Base 0.65 0.80 - 0.00 缺乏领域知识注入，逻辑推理空泛，多为通用性描述. RAG-黔西南 0.89 0.81 0.90 0.95 论据充分，能够实现从结论到原始本地文献的精准溯源. 注：通用模型与基座模型未挂载本地知识库，故“上下文精确度”不适用；GeoGPT-R1-Preview测试时挂载通用地学库但未包含本文构建的知识库.
下载: 导出CSV
表 4 基于知识图谱的中国黔西南‒美国内华达卡林型金矿关键成矿要素对比
Table 4. Knowledge graph-based comparison of key metallogenic factors of Carlin-type gold deposits in Southwest Guizhou of China and Nevada of the USA
对比维度 中国黔西南矿集区 美国内华达矿集区 对比分析 构造背景 华南板块（被动大陆边缘） 北美板块（被动大陆边缘） 二者均形成于伸展背景，但黔西南的板内裂谷环境更为突出. 控矿构造 以NE向、NNE向高角度断裂为主（如普安‒马场断裂），褶皱（如清水河背斜）控矿作用显著. 以NNW向高角度断裂为主（如Getchell断裂），低角度拆离断层（如Roberts山拆离断层）作用关键. 中国黔西南呈现“断‒褶”双重控制，而美国内华达的“断‒断”（高角度与低角度）组合更典型. 赋矿地层 泥盆系（D）至二叠系（P），以泥质岩、粉砂岩（如龙潭组）为绝对主导. 寒武系（∈）至泥盆系（D），以碳酸盐岩（如Popovich组）为主. 赋矿围岩的岩性差异（硅质‒泥质vs碳酸盐岩）是导致二者地球化学特征差异的根本原因. 蚀变矿物 硅化、黄铁矿化、雄黄‒雌黄矿化、黏土化（高岭石、伊利石）. 硅化、黄铁矿化、雄黄‒雌黄矿化、重晶石化、碳酸盐化. 中国黔西南的“黏土化”更为发育，可能与含泥质围岩有关；美国内华达的“重晶石化”独有. 矿石矿物 毒砂、黄铁矿、雄黄、雌黄 毒砂、黄铁矿、雄黄、雌黄 组合高度相似，均以含砷矿物为主要载金矿物.
下载: 导出CSV
表 5 右江盆地卡林型金矿的图谱聚类特征与成矿系列划分
Table 5. Clustering characteristics and metallogenic system division of Carlin-type gold deposits in Youjiang basin based on knowledge graph
成矿系列划分 图谱聚类簇组合 关键控矿要素 地质成因机制 Ⅰ类断裂‒构造主控系列 烂泥沟‒丫他‒华新‒大沟‒百地‒塘新寨‒那郎 高角度断裂、毒砂/黄铁矿化 属原生深部断控系统.受区域性深大断裂及其次级破碎带严格控制，矿体呈陡倾状延伸，切穿地层能力强，完整保留了深部含矿流体沿构造通道垂向运移与沉淀的记录. 坝赖‒弄朗 高角度逆冲断层、断裂破碎带、许满组复理石 属碎屑岩容矿的断控型亚类.受F1、F2高角度断层破碎带控制，具有显著的切层特征，属于深部流体在碎屑岩区沿断裂通道贯入成矿. Ⅱ类层位‒岩性主控系列 泥堡‒戈塘 不整合面、硅化顶盖、层间滑脱带 属不整合面‒层控型亚类.矿体受上覆不透水层（如黏土岩）的岩性封闭制约，成矿流体沿不整合面侧向运移特征显著，发育大规模的层状硅化体. 紫木凼 背斜构造、Hg‒Tl(铊)异常、雄黄矿化 属背斜‒层控型亚类.位于灰家堡背斜核部，受层间虚脱空间与张性裂隙控制，矿体多呈顺层或鞍状产出，且表现出对Hg-Tl元素的特殊地球化学亲和性. Ⅲ类表生风化‒堆积系列 万人洞‒上大观 第四系黏土、茅口组古侵蚀面、褐铁矿化 属红土‒岩溶堆积型成矿系统.严格受控于茅口组古岩溶侵蚀面与第四系风化壳，矿体赋存于岩溶洼地或剥夷面之上的红土黏土及角砾岩中，表征为原生矿体经强烈的表生氧化、崩塌堆积与次生富集. 油菜冲‒虎场 残坡积物、褐铁矿化、黏土化 老万场‒豹子洞 岩溶塌陷堆积、红土化、游离金
下载: 导出CSV
-
Cao, S. T., Hu, R. Z., Zhou, Y. Z., et al., 2026. Analysis of Trajectories and Developmental Prospects of Research on Carlin-Type Gold Deposits on the Basis of Big Data Community Detection Algorithms. Ore Geology Reviews, 188: 106989. https://doi.org/10.1016/j.oregeorev.2025.106989 Chen, W. B., Wei, B. G., Yang, T. C., et al., 2009. Geological Character and Prospecting Potential of Nibao Gold Deposit in Pu'an County, Guizhou. Guizhou Geology, 26(3): 170-176 (in Chinese with English abstract). Cheng, Q. M., 2025. A New Paradigm for Mineral Resource Prediction Based on Human Intelligence-Artificial Intelligence Integration. Earth Science Frontiers, 32(4): 1-19 (in Chinese with English abstract). Deng, J., Wang, Q. F., 2016. Gold Mineralization in China: Metallogenic Provinces, Deposit Types and Tectonic Framework. Gondwana Research, 36: 219-274. https://doi.org/10.1016/j.gr.2015.10.003 Dong, Y. H., Wang, Y. Z., Tian, J. T., et al., 2025. Research Progress on Porphyry Copper Deposit Prediction Based on Knowledge Graphs. Earth Science Frontiers, 32(4): 280-290 (in Chinese with English abstract). Feng, T. T., Cai, S. R., Zhang, Z. J., 2025. Mining Elements of Carbonatite-Type Rare Earth Deposits Based on Knowledge Map. Earth Science Frontiers, 32(4): 262-279 (in Chinese with English abstract). Fu, Y., Wang, M. G., Wang, C. B., et al., 2025. GeoMinLM: A Large Language Model in Geology and Mineral Survey in Yunnan Province. Ore Geology Reviews, 182: 106638. https://doi.org/10.1016/j.oregeorev.2025.106638 Goldfarb, R., Qiu, K., Deng, J., et al., 2019. Orogenic Gold Deposits of China. Geological Society, London, Special Publications, 480: 263-288. https://doi.org/10.1144/SP480-2018-175 Hofstra, A. H., Christensen, O. D., 2002. Comparison of Carlin-Type Au Deposits in the United States, China, and Indonesia-Implications for Genetic Models and Exploration. US Geological Survey Open-File Report, 2-131. Hofstra, A. H., Cline, J. S., 2000. Characteristics and Models for Carlin-Type Gold Deposits. Reviews in Economic Geology, 13: 163-220. https://doi.org/10.5382/Rev.13.05 Hu, R. Z., Fu, S. L., Huang, Y., et al., 2017. The Giant South China Mesozoic Low-Temperature Metallogenic Domain: Reviews and a New Geodynamic Model. Journal of Asian Earth Sciences, 137: 9-34. https://doi.org/10.1016/j.jseaes.2016.10.016 Hu, Y. J., Mai, G. C., Cundy, C., et al., 2023. Geo-Knowledge-Guided GPT Models Improve the Extraction of Location Descriptions from Disaster-Related Social Media Messages. International Journal of Geographical Information Science, 37(11): 2289-2318. https://doi.org/10.1080/13658816.2023.2266495 Jiang, Z. Y., Zhong, L., Sun, M. S., et al., 2024. Efficient Knowledge Infusion via KG-LLM Alignment. arXiv, 2406.03746. https://arxiv.org/abs/2406.03746 Lewis, P., Perez, E., Piktus, A., et al., 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. arXiv, 2005.11401. https://arxiv.org/abs/2005.11401 Li, B. W., Wang, Y. Z., Ding, Z. J., et al., 2025. Intelligent Search Technology for Jiaodong Gold Deposits Based on Large Models and GraphRAG. Earth Science Frontiers, 32(4): 155-164 (in Chinese with English abstract). Li, G. Z., Wang, P., Ke, W. J., 2023. Revisiting Large Language Models as Zero-Shot Relation Extractors. arXiv, 2310.05028. https://arxiv.org/abs/2310.05028 Ma, X. G., 2022. Knowledge Graph Construction and Application in Geosciences: A Review. Computers & Geosciences, 161: 105082. https://doi.org/10.1016/j.cageo.2022.105082 Mao, B. J., Ran, R. D., Kuang, S. D., et al., 2018. Genesis of the Getang Gold Deposit in the Southwest Guizhou. Contributions to Geology and Mineral Resources Research, 33(2): 168-175 (in Chinese with English abstract). Peng, J. J., Lin, K., 2024. Knowledge Graph Analysis of Mineralization Laws Research of Lithium Ore. China Mining Magazine, 33(9): 228-235 (in Chinese with English abstract). Pi, Q. H., Hu, R. Z., Xiong, B., et al., 2017. In Situ SIMS U-Pb Dating of Hydrothermal Rutile: Reliable Age for the Zhesang Carlin-Type Gold Deposit in the Golden Triangle Region, SW China. Mineralium Deposita, 52(8): 1179-1190. https://doi.org/10.1007/s00126-017-0715-y Qiu, Q. J., Tian, M., Wu, Q. R., et al., 2025. Construction and Application of Geological Knowledge Graph Based on Multi-Source Heterogeneous Data. Earth Science Frontiers, Online. (in Chinese with English abstract). https://doi.org/10.13745/j.esf.sf.2024.11.69 Raiaan, M. A. K., Mukta, M. S. H., Fatema, K., et al., 2024. A Review on Large Language Models: Architectures, Applications, Taxonomies, Open Issues and Challenges. IEEE Access, 12: 26839-26874. https://doi.org/10.1109/ACCESS.2024.3365742 Ran, R. D., 2005. Characteristic and Metallogenic Mechanism of the Gold Deposits with Karst Structure as Holding Ore Space in the Southwest of Guizhou-Taking the Getang Gold Deposit in Anlong as an Example. Guizhou Geology, 22(1): 14-21 (in Chinese with English abstract). Shi, L. Y., Zuo, R. G., 2026. Foundation Model for Mineral Prospectivity Mapping. Earth Science, 51(3): (in Chinese with English abstract). Sutanto, P., Santoso, J., Setiawan, E. I., et al., 2024. LLM Distillation for Efficient Few-Shot Multiple Choice Question Answering. arXiv, 2412.09807. https://arxiv.org/abs/2412.09807 Tan, S. M., Shi, G. D., Lei, L. Q., et al., 2007. Carlin-Type Gold Deposits Distribution and Prospecting in China. Geological Survey and Research, 30(4): 289-294 (in Chinese with English abstract). Tao, P., Li, P. G., Li, K. Q., 2002. The Structure of the Deposits of the Nibao Goldfield and Its Relationship with Metallogenesis. Guizhou Geology, 19(4): 221-227 (in Chinese with English abstract). Tian, S. Y., Luo, Y. Y., Xu, T. Z., et al., 2024. KG-Adapter: Enabling Knowledge Graph Integration in Large Language Models through Parameter-Efficient Fine-Tuning. Annual Meeting of the Association for Computational Linguistics, Bangkok. https://doi.org/10.18653/v1/2024.findings-acl.229 Wang, C. B., Wang, M. G., Wang, B., et al., 2024. Knowledge Graph-Infused Quantitative Mineral Resource Forecasting. Earth Science Frontiers, 31(4): 26-36 (in Chinese with English abstract). Wang, Q. F., Groves, D., 2018. Carlin-Style Gold Deposits, Youjiang Basin, China: Tectono-Thermal and Structural Analogues of the Carlin-Type Gold Deposits, Nevada, USA. Mineralium Deposita, 53(7): 909-918. https://doi.org/10.1007/s00126-018-0837-x Xie, Z. J., Xia, Y., Cline, J. S., et al., 2018. Are There Carlin-Type Gold Deposits in China? A Comparison of the Guizhou, China, Deposits with Nevada, USA, Deposits. Reviews in Economic Geology, 20: 187-233. https://doi.org/10.5382/rev.20.06 Xie, Z. J., Xia, Y., Cline, J., et al., 2019. A Comparison between Carlin-Type Au Deposits in Guizhou of China and Nevada of the USA and Its Implications for Exploration. Mineral Deposits, 38(5): 1077-1093 (in Chinese with English abstract). Yang, X., Sun, L., Liu, M. L., et al., 2025. Knowledge Graph Construction with BERT-BiLSTM-IDCNN-CRF and Graph Algorithms for Metallogenic Pattern Discovery: A Case Study of Pegmatite-Type Lithium Deposits in China. Ore Geology Reviews, 179: 106514. https://doi.org/10.1016/j.oregeorev.2025.106514 Zhang, B. W., Soh, H., 2024. Extract, Define, Canonicalize: An LLM-Based Framework for Knowledge Graph Construction. arXiv, 2404.03868. https://arxiv.org/abs/2404.03868 Zhang, B. Y., Tang, J. C., Zhang, T. Y., et al., 2026. Knowledge Graph and Question-Answering Model for Geological Prospecting Empowered by Large Language Models. Earth Science, 51(3): 982-995 (in Chinese with English abstract). Zhang, Y. F., Wei, C., He, Z. T., et al., 2024. GeoGPT: An Assistant for Understanding and Processing Geospatial Tasks. International Journal of Applied Earth Observation and Geoinformation, 131: 103976. https://doi.org/10.1016/j.jag.2024.103976 Zhang, Z. J., Yang, Z. X., Jian, F. Y., et al., 2025. Interpretability-Enhanced Mineral Prospectivity Models: A Synergistic Approach Using Large Language Models, Knowledge Graphs, and Machine Learning. Mathematical Geosciences, Online. https://doi.org/10.1007/s11004-025-10231-3 Zhao, M. L., Zhang, Z. J., Yang, J., et al., 2025. Knowledge Graph Construction and Knowledge Discovery for Porphyry Copper Deposits. Ore Geology Reviews, 186: 106875. https://doi.org/10.1016/j.oregeorev.2025.106875 Zhao, P. D., 2019. Characteristics and Rational Utilization of Geological Big Data. Earth Science Frontiers, 26(4): 1-5 (in Chinese with English abstract). Zhou, Y. Z., Zhang, Q. L., Huang, Y. J., et al., 2021a. Constructing Knowledge Graph for the Porphyry Copper Deposit in the Qingzhou-Hangzhou Bay Area: Insight into Knowledge Graph Based Mineral Resource Prediction and Evaluation. Earth Science Frontiers, 28(3): 67-75 (in Chinese with English abstract). Zhou, Y. Z., Zuo, R. G., Liu, G., et al., 2021b. The Great-Leap-Forward Development of Mathematical Geoscience during 2010-2019: Big Data and Artificial Intelligence Algorithm Are Changing Mathematical Geoscience. Bulletin of Mineralogy, Petrology and Geochemistry, 40(3): 556-573, 777 (in Chinese with English abstract). 陈文斌, 韦标根, 杨天才, 等, 2009. 贵州普安县泥堡金矿床地质特征与找矿潜力. 贵州地质, 26(3): 170-176. 成秋明, 2025. 面向人类智能与人工智能融合的矿产资源预测新范式. 地学前缘, 32(4): 1-19. 董宇浩, 王永志, 田江涛, 等, 2025. 基于知识图谱的斑岩型铜矿预测研究进展. 地学前缘, 32(4): 280-290. 冯婷婷, 蔡诗柔, 张振杰, 2025. 基于知识图谱的碳酸岩型稀土矿成矿要素挖掘. 地学前缘, 32(4): 262-279. 李博文, 王永志, 丁正江, 等, 2025. 基于大模型与GraphRAG的胶东金矿智能搜索技术. 地学前缘, 32(4): 155-164. 毛彬吉, 冉瑞德, 况顺达, 等, 2018. 黔西南戈塘金矿成因再认识. 地质找矿论丛, 33(2): 168-175. 彭晶晶, 林锴, 2024. 锂矿成矿规律研究的知识图谱分析. 中国矿业, 33(9): 228-235. 邱芹军, 田苗, 吴麒瑞, 等, 2025. 基于多源异构数据的地质知识图谱构建与应用. 地学前缘, 知网首发. https://doi.org/10.13745/j.esf.sf.2024.11.69 冉瑞德, 2005. 黔西南岩溶构造容矿金矿床特征及成矿机理: 以安龙戈塘金矿床为例. 贵州地质, 22(1): 14-21. 师路易, 左仁广, 2026. 矿产预测大模型. 地球科学, 51(3): . doi: 10.3799/dqkx.2025.190 谭仕敏, 施国栋, 雷良奇, 等, 2007. 中国卡林型金矿的分布规律及找矿前景. 地质调查与研究, 30(4): 289-294. 陶平, 李沛刚, 李克庆, 2002. 贵州泥堡金矿区矿床构造及其与成矿的关系. 贵州地质, 19(4): 221-227. 王成彬, 王明果, 王博, 等, 2024. 融合知识图谱的矿产资源定量预测. 地学前缘, 31(4): 26-36. 谢卓君, 夏勇, Cline, J. S., 等, 2019. 中国贵州与美国内华达卡林型金矿对比及对找矿勘查的指示作用. 矿床地质, 38(5): 1077-1093. 张宝一, 唐嘉成, 张彤蕴, 等, 2026. 大语言模型赋能的地质找矿知识图谱与问答模型构建. 地球科学, 51(3): 982-995. 赵鹏大, 2019. 地质大数据特点及其合理开发利用. 地学前缘, 26(4): 1-5. 周永章, 张前龙, 黄永健, 等, 2021a. 钦杭成矿带斑岩铜矿知识图谱构建及应用展望. 地学前缘, 28(3): 67-75. 周永章, 左仁广, 刘刚, 等, 2021b. 数学地球科学跨越发展的十年: 大数据、人工智能算法正在改变地质学. 矿物岩石地球化学通报, 40(3): 556-573, 777. -
点击查看大图
计量
- 文章访问数: 224
- HTML全文浏览量: 54
- PDF下载量: 61
- 被引次数: 0
