Knowledge acquisition from chemical accident databases using an ontology-based method and natural language processing

Abstract

Accident databases are used to learn from past accidents and avoid future accidents in the chemical process industry. Classical accident databases can be tedious to use because the database entries were written over the years by various persons in different styles. Thus, accident case entries must be interpreted to identify cause-effect relationships and recognize lessons learned. Semantically enriched accident databases can make information retrieval more efficient. In this research approach Natural Language Processing methods are used to extract information from a chemical accident database. Additionally, substance information is enriched using web scraping techniques. Afterwards, a predefined ontology structure is automatically populated with the extracted information. The ontology-based chemical accident database provides additional accident exploration capabilities that can be used by human experts or computer systems. The results indicate that the proposed extraction method is well suited to extract accident information. The ontology is useful to discover causal accident relations due to the semantically described context of accident information. The method described can be adapted to other databases with minor adaptations and refinements. By combining various ontology-based accident databases a human and machine-processable, and sharable knowledge structure can be provided to reuse knowledge across companies and countries.