An improved text mining approach to extract safety risk factors from construction accident reports

Workplace accidents in construction commonly cause fatal injury and fatality, resulting in economic loss and negative social impact. Analysing accident description reports helps identify typical construction safety risk factors, which then becomes part of the domain knowledge to guide safety management in the future. Currently, such practice relies on domain experts' judgment, which is subjective and time-consuming. This paper developed an improved approach to identify safety risk factors from a volume of construction accident reports using text mining (TM) technology. A TM framework was devised, and a workflow for building a tailored domain lexicon was established. An information entropy weighted term frequency (TF H) was proposed for term-importance evaluation, and an accumulative TF H was proposed for threshold division. A case study of metro construction projects in China was conducted. A list of 37 safety risk factors was extracted from 221 metro construction accident reports. The result shows that the proposed TF H approach performs well to extract important factors from accident reports, solving the impact of different report lengths. Additionally, the obtained risk factors depict critical causes contributing most to metro construction accidents in China. Decision-makers and safety experts can use these factors and their importance degree while identifying safety factors for the project to be constructed.

Automated summary

This study aims to identify safety risk factors in construction accident reports using text mining technology, proposing an information entropy weighted term frequency (TF H) method for term importance evaluation. The results show that this approach effectively extracts important factors and addresses the impact of different report lengths, depicting critical causes contributing most to metro construction accidents in China.