Knowledge graph-driven decision support for cross-regional solid mineral deposits

This is the authors' abstract. We don't add key points for this paper.

Introduction To tackle the challenges of fragmented multi-regional data and obscured correlations in solid mineral deposit management, this study proposes a novel knowledge graph construction and application framework centered on a Content-Position Attention (CPA) mechanism. Methods The core CPA model features a content-position fusion encoder and an entity grid decoder, which work in tandem to capture textual context and entity positional information, enabling precise extraction of overlapping and nested entity-relation triples. Based on this model, we built a domain ontology and integrated it with the Neo4j graph database to create a comprehensive knowledge graph for solid mineral deposits in South China. Results Evaluated on the self-constructed SC-Mineral dataset, the CPA model attained an overall F1-score of 91.8%. In complex scenarios including entity-pair overlap, single-entity overlap, and subject-object nested overlap, the model obtained F1-scores of 92.8%, 92.5%, and 89.8% respectively, demonstrating excellent capability in handling complex relations. Discussion The resulting knowledge graph system enables intelligent information retrieval, multi-dimensional correlation analysis, and mineralization potential prediction, demonstrating its practical effectiveness in supporting cross-regional resource management and intelligent decision-making. This work offers a replicable technical pathway for knowledge management and decision analysis in the mineral resources domain.