Data-Driven Deep Insights into Mineral Systems Using Knowledge Graphs

Knowledge graphs (KGs) are a powerful tool in big data analytics， capable of storing， retrieving， and analyzing diverse textual data while revealing complex correlations. Mineral systems represent intricate networks of interrelated geodynamic events characterized by varied survey data， such as geological maps and geochemical compositions. Therefore， constructing structural models of these systems is challenging due to the extensive and heterogeneous nature of the data. To address this， we employed KGs to integrate and extract survey data， enhancing our understanding of mineral systems. In our case study， we utilized a deep learning model (BERT-BiLSTM-CRF)， which amalgamates bidirectional encoder representations from transformers (BERT)， bidirectional long-short-term memory (BiLSTM)， and conditional random field (CRF)， to identify and extract entities and relationships from text associated with magmatic-hydrothermal deposits in the Eastern Tianshan orogenic belt. This enabled us to construct a KG of these mineral systems and analyzed their centrality， similarity， and community structures. Our findings demonstrate that the BERT-BiLSTM-CRF model effectively automats the extraction of entities and relationships from unstructured text， serving as a valuable tool for building KGs of mineral systems and providing deeper， data-driven insights into their complexities. This approach significantly enhances our comprehension of geodynamic conditions and ore-forming processes in the region.