Domain-knowledge guidance and prompt engineering for extracting geologic entity relationship

By harmonizing multi-source heterogeneous geoscience data, the geological knowledge graph enables rapid and precise discovery of latent knowledge. Geologic entity relationship extraction is a key technology in the construction of knowledge graphs. To address the specialization and complexity of geological texts and the problems of data imbalance and small size of the dataset, this paper proposes a relation extraction model that integrates domain knowledge guidance with prompt engineering. Data augmentation is performed using prompt learning, and manually constructed multi-prompt templates and label mapping strategies for fusing geologic domain relations and descriptions are designed to integrate geologic domain features. Comprehensive experiments demonstrate that the proposed model achieves 65.5 % precision and 64 % F1-score on geological entity-relation extraction, yielding a measurable performance gain and furnishing a viable technical cornerstone for large-scale geological knowledge-graph construction.