How to realize the knowledge reuse and sharing from accident reports? A knowledge-driven modeling method combining ontology and deep learning

Abstract

The exploration and understanding of past accidents are of great significance in enhancing the process safety. However, manually reading and analyzing a large number of accident reports is a time-consuming and inefficient task. In this study, a novel modeling method is developed to build the knowledge graph of process safety accidents, aiming to overcome the problem of knowledge reuse and sharing. Firstly, the dataset consists of 409 process safety accident reports selected from the official website of the Ministry of Emergency Management of China. Secondly, the ontology design schema is defined based on the seven-step method, including 34 ontology classes and 11 relations. Then, a new joint extraction model for the process domain is proposed based on the CasRel framework, which achieves 95.85% in precision, 61.54% in recall, and 74.95% in F₁-score. Finally, the knowledge graph containing 9192 nodes and 11,257 edges is constructed in the Neo4j graph database, followed by the discussion of various related applications such as query, statistics, and analysis. The results indicate that the proposed method is a useful tool for obtaining valuable knowledge from accident reports, contributing to analysis and prevention of accidents.