A data-driven narratives skeleton pattern recognition from accident reports dataset for human-and-organizational-factors analysis
Shuo Yang, Micaela Demichela
Abstract
Accidents in the process industry involve several interacting factors, including human and organizational factors (HOFs). A long-standing obstacle to HOFs analysis is lack of data. Accident reports are an essential data source to learn from the past and contain HOFs-related data, but they are usually unstructured text in a not standardized format. Some studies have explored the extraction of information automatically from accident reports based on Natural Language Processing (NLP) techniques. However, they were not dedicated to HOFs. Risk communication is considered an essential pillar in safety and risk science. This research develops a HOFs-focused risk communication framework based on the NLP techniques that can support risk assessment and mitigation. The proposed approach automatically extracts the target groups oriented “Who, When, Where, Why” (4Ws) information from accident reports. This framework was applied to explore the eMARS database. The results show that the “4Ws” skeleton of narratives has appreciated performance in pattern recognition and holistic information analysis. The graphical representation interfaces are designed to display the features of HOFs-related accidents, which can better be communicated to the sharp-end operators and decision-makers.