A framework for process risk assessment incorporating prior hazard information in text mining models using chunking

Journal article that combines domain hazard knowledge with NLP/text mining to extract ‘chunks’ from incident descriptions representing faults and process safety events, then applies unsupervised/semi-supervised learning to reconstruct chains of events and generate fault trees. Demonstrated on an industrial steel plant dataset with reported agreement against HSE expert inputs. For PSM, it enables systematic learning from narratives in incident/near-miss reports, supporting hazard identification, precursor discovery, and prioritization of risk-reduction actions—especially valuable when organizations have large backlogs of unstructured safety reports.