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

Abstract

Compared to traditional geological data processing methods, knowledge graphs are more effective in calculating and processing the associated information and implicit geological knowledge within the data, helping to accurately grasp the underlying patterns and relationships of geological phenomena. To further optimize the semantic representation of geological text data and extract more detailed feature information, this study introduces the dilated convolutional neural network (IDCNN) layer into the Bert-BiLSTM-CRF model, constructing the Bert-BiLSTM-IDCNN-CRF framework for the precise extraction of lithium deposit named entities.This framework is then used to construct a knowledge graph for granite (pegmatite) lithium deposits in China. Experimental results demonstrate that the Bert-BiLSTM-IDCNN-CRF model exhibits excellent performance in processing Chinese geological text data, achieving a precision of 89%, a recall rate of 87%, and an F1 score of 88%. These results confirm the model's high effectiveness in geological named entity recognition and extraction tasks. Based on this, the study further employs centrality and similarity algorithms from graph theory to deeply analyze the metallogenic characteristics and potential patterns of lithium deposits. This analysis successfully identifies key influencing factors and core nodes for each lithium belt, providing a solid scientific foundation for subsequent lithium exploration target area delineation.