Natural language processing tool for identifying influencing factors in human reliability analysis and summarizing accident reports

Tools

Johnson, Karl and Patelli, Edoardo and Morais, Caroline; Brito, Mário P. and Aven, Terje and Baraldi, Piero and Čepin, Marko and Zio, Enrico, eds. (2023) Natural language processing tool for identifying influencing factors in human reliability analysis and summarizing accident reports. In: 33rd European Safety and Reliability Conference (ESREL 2023). Research Publishing, Singapore. ISBN 9789811880711 (https://doi.org/10.3850/978-981-18-8071-1_P294-cd)

[thumbnail of Johnson-etal-ESREL-2023-Natural-language-processing-tool-for-identifying]
Preview
 Text. Filename: Johnson-etal-ESREL-2023-Natural-language-processing-tool-for-identifying.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (1MB)| Preview

Abstract

The development of a tool based on Natural Language Processing (NLP) models is presented. The presented tool is an improvement on the original virtual human factors classificator developed to assist experts with extracting the organizational, technological, and individual factors that may trigger human errors. To identify the performance shaping factors, the approach proposed is based on classifying text according to previously labelled accident reports by human experts, making use of BERT (Bidirectional Encoder Representations from Transformers), a popular transformer-based machine learning model for NLP. In addition, a method to provide a summarization of each accident report is presented. This provides further detailed context alongside with the identified performance shaping factors, without the need of reading the entire report which is generally a significant task. The tool performs abstractive summarization as it aims to understand the entire report and generate paraphrased text to summarize the main points. In this work, BART (Bidirectional and Auto-Regressive Transformers), which is a denoising autoencoder for pre-training sequence-to-sequence models, has been used as the basis for the text summarization model.

ORCID iDs

Johnson, Karl, Patelli, Edoardo ORCID logoORCID: https://orcid.org/0000-0002-5007-7247 and Morais, Caroline; Brito, Mário P., Aven, Terje, Baraldi, Piero, Čepin, Marko and Zio, Enrico
CORE (COnnecting REpositories)