Answers are grounded in this paper and cite their sources.
Purpose To develop a temporal knowledge graph framework to overcome the limitations of static knowledge representation and support risk evolution analysis in metro construction accidents. Design/methodology/approach A dual-layer knowledge extraction framework was constructed. Bidirectional encoder representations from transformers (BERT)–bidirectional long short-term memory–conditional random field was used to extract accident case knowledge, while BERT + CasRel integrated with a temporal extraction layer was employed to extract time-aware quadruples of accident-process knowledge. Based on 575 accident reports, the extracted knowledge was organized into a temporal knowledge graph and further used for accident timeline visualization, risk evolution path analysis, and multidimensional risk chain identification. Findings Compared with the rule-based temporal extraction model, the proposed BERT + CasRel + T model improved temporal alignment accuracy and Quadruple F1 by 13.4 and 4.2 percentage points, respectively. In the case application of Xuzhou Metro Line 6, the expert evaluation score for the identified risk evolution paths reached 4.52 out of 5.00. Originality/value This study structures metro construction accident knowledge into time-aware quadruples and integrates entity recognition, relation extraction, and temporal alignment into a domain-specific extraction framework. The resulting temporal knowledge graph supports process-oriented representation and risk evolution analysis from accident reconstruction to risk-chain reasoning.
No takes yet. Share an insight, caveat, or question.
Guo et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: