Investigation and validation of a novel text mining methodology for occupational accident analysis and prevention

This study aimed to investigate and validate a novel text mining methodology for occupational accident analysis and prevention. To achieve this goal, the credibility, internal validity, external validity and reliability of a new text mining approach was examined with respect to occupational accident databases. Three successive studies were conducted: Study I performed a trial text mining analysis of a small database by using the default settings of SAS Text Miner. Studies II and III applied a novel text mining methodology to a small-size and a large-size database respectively to validate its credibility, internal validity, external validity and reliability. Examination of these four areas enables the user to evaluate the application of text mining techniques in accident analysis, to identify the strength and limitations of text mining techniques, and to provide recommendations for OHS professions. It appears that the true value of a text mining methodology becomes most apparent as the sample size of the dataset increases, based on a positive relationship between the value of the software to an organisation and the size of the datasets that these larger organisations will tend to have access. The adoption of text mining analysis is probably most feasible for large organisations such as multinational corporations, government departments and workplace regulatory authorities, as it is these organisations that can more easily absorb the labour-intensive steps required to conduct the most meaningful text mining analysis of occupational injury data.