A deep learning-based method for deep information extraction from multimodal data for geological reports to support geological knowledge graph construction

Earth science research has entered a period of major transition centered on building new knowledge systems and driven by the overwhelming availability of vast amounts of data. Obtaining a comprehensive understanding of earth processes can be challenging due to the complexity and diversity of the various geological data. To tackle this issue, the paper proposes adopting data-driven knowledge discovery techniques for analyzing mineral exportation reports. We employed natural language processing and text mining, image segmentation, and deep neural networks to extract the geological entity and topic information, understand the geological map object associations, and recognize geological table element associations linked to mineralization to support mineral exploration with a set of mineral exploration reports. The experimental results demonstrate the following: (1) extracting expert knowledge from mineral exploration texts can further enrich the geological information of the region; (2) recognizing the content, semantic and attribute information of geological objects from geological maps and tables, is important for understanding the deposits, geological mineralization associations, hidden geological rules in the region; and (3) constructing a geological knowledge graph or knowledge base for mineral reports can provide significant information for further mineral exploration.