Analysis of Occupational Accident Narratives in the Manufacturing Sector Using Latent Dirichlet Allocation

Despite advances and technological innovations in the field of occupational safety, workplace accidents remain a major economic and social problem. New risks emerge as work processes become more complex, and this affects employee health, productivity, and economic sustainability. As a result, the study's goal is to investigate sectoral errors that contribute to occupational accidents in the manufacturing sector. In this context, approximately 1,800,000 occupational accident descriptions obtained from the Social Security Institution (SGK) were analyzed using natural language processing and text mining techniques to identify patterns in accidents, determine thematic structures, and thus gain a better understanding of the underlying root causes. In this context, pre-processing steps (text cleaning, root finding, lemmatization), and Latent Dirichlet Allocation (LDA) method were applied, and hidden structures were revealed with unsupervised learning algorithms. The findings enable the evaluation of occupational accidents in narrative-based dimensions and provide new perspectives for decision-makers in developing preventive strategies for errors made on a sectoral basis.