Scholarlink

<p>Developing interpretable predictive models for mineral prospectivity remains a persistent challenge in geoscience, as traditional data-driven methods often lack transparency, limiting geological understanding and validation of predictions. This study introduces an approach that integrates large language models (LLMs) with text data to construct a knowledge graph (KG), which is subsequently mined to identify and select predictive variables related to specific ore deposit types. Based on the extracted knowledge, an interpretable prediction framework (KG-SA-GC) is developed by combining S-A fractal filtering (SA) with the gcForest machine learning algorithm (GC). The framework is applied to the Nanling region of South China as a case study, successfully identifying key geochemical predictors associated with rare metal mineralization. The results demonstrate the potential of utilizing LLMs and KGs to build interpretable, accurate, and scalable models for mineral prospectivity, highlighting the broader applicability of artificial intelligence (AI)-driven methodologies in advancing mineral exploration across diverse geological settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Interpretability-Enhanced Mineral Prospectivity Models: A Synergistic Approach Using Large Language Models, Knowledge Graphs, and Machine Learning

  • 期刊论文

  • 发表日期：

    2025年9月17日

查看全文
  • Zhen-Jie Zhang,
  • Zi-Xing Yang,
  • Fu-Yuan Jian,
  • Si-Yuan Ban,
  • Yi-Ming Wang

摘要

Developing interpretable predictive models for mineral prospectivity remains a persistent challenge in geoscience, as traditional data-driven methods often lack transparency, limiting geological understanding and validation of predictions. This study introduces an approach that integrates large language models (LLMs) with text data to construct a knowledge graph (KG), which is subsequently mined to identify and select predictive variables related to specific ore deposit types. Based on the extracted knowledge, an interpretable prediction framework (KG-SA-GC) is developed by combining S-A fractal filtering (SA) with the gcForest machine learning algorithm (GC). The framework is applied to the Nanling region of South China as a case study, successfully identifying key geochemical predictors associated with rare metal mineralization. The results demonstrate the potential of utilizing LLMs and KGs to build interpretable, accurate, and scalable models for mineral prospectivity, highlighting the broader applicability of artificial intelligence (AI)-driven methodologies in advancing mineral exploration across diverse geological settings.