Mapping a Knowledge Graph of Flooding in Academic Literature Through Full‐Text Entity Extraction

ABSTRACT Academic literature with long‐tail characteristics contains rich knowledge resources but is not easily discovered through limited manual knowledge extraction capabilities. This study proposed a refined extraction strategy, transitioning from “full text to sentence,” in the field of flood disaster risk reduction. Sentences describing research methods were identified from the full texts of 5180 articles published between 1990 and 2020. Research method entities—including algorithms, software, and data—were extracted using optimal deep learning models. A flooding knowledge graph was constructed and applied to several flood control scenarios. The results showed that the BiLSTM‐CRF model outperformed more complex alternatives. In all, 2144 research methods, 291 software tools, and six types of remote sensing data sources were obtained based on extracted usage method sentences. The flooding knowledge graph contained 42,420 nodes and 78,242 edges. The proposed refined knowledge entity extraction method provides a reference for related knowledge graph mapping based on big data.