Using Natural Language Processing to Generate Risk Assessment Checklists From Workplace Descriptions

Workplace risk assessments are mandatory in Germany. Despite this, up to 50% of workplaces are estimated to never have been assessed. We propose to empower occupational safety practitioners by employing artificial intelligence techniques. From previous work we gathered that most of the available risk assessment data is more or less digitalized and more or less structured text. Therefore, we focus the algorithm family of natural language processing to boost the capabilities of analysing large amounts of documents. More specifically, we show the feasibility of retraining an algorithm for handling accident data and hazard identification and risk assessment checklists as well as the transferability to a German text corpus. A dataset of roughly 2500 job descriptions in German forms the basis for this German text corpus. We further explore how the interaction with such a system should be designed to accommodate for the specifics of interacting with artificial intelligence as well as maintaining the users' competencies.

Keywords: Risk assessment, Natural language processing, Occupational safety, Workplace, Artificial intelligence, Hazard identification.

Download PDF