Accident Case Retrieval and Analyses: Using Natural Language Processing in the Construction Industry

Abstract

Knowledge management for construction accident cases can identify dangerous conditions and prevent accidents by controlling risks on-site. However, because accident cases are recorded as unstructured text data, significant time and effort are required to retrieve and analyze the knowledge a user wants. To overcome these limitations, this research proposes a knowledge management system for construction accident cases using natural language processing. For this purpose, two models were developed that can retrieve appropriate cases according to user intentions and automatically analyze tacit knowledge from construction accident cases. In the retrieval model, the query is expanded using a construction accident case thesaurus. Ranking is calculated using Okapi BM25 and weighting according to the thesaurus. In the analysis model, knowledge is automatically extracted using rule-based and conditional random field (CRF) methods. The proposed system can retrieve results that are 97% relevant to the accident cases the user intended and can automatically analyze knowledge with accuracies of 93.75% and 84.13% for the rule-based and CRF models, respectively. The results demonstrate the potential of knowledge discovery from accident reports for more-effective safety management.