Uncovering Key Construction Injury Causal Factors Using Topic Modeling and Word Co-Occurrence Network Analysis Methods

Abstract The construction industry suffers from a high rate of injuries, with unstructured textual injury reports containing rich but underutilized causal information. Traditional analytical methods struggle with the high dimensionality and semantic sparsity of this data, limiting the depth of causal insight. To address this gap, this research develops an analytical method that combines latent Dirichlet allocation (LDA) topic modeling and word co-occurrence network (WCN) analysis. Applied to 25,980 real-world injury reports, the proposed method first identified three coherent injury topics. Subsequently, WCN analysis with centrality metrics (betweenness, closeness) revealed not only explicit key factors but also situation-dependent implicit factors that act as critical intermediaries in specific injury pathways. Crucially, the validity and industry relevance of the findings were systematically benchmarked against the US Occupational Injury and Illness Classification System (OIICS). The contributions of this research are as follow: (1) Methodologically, it advances current analytical practices by developing and demonstrating an integrated framework that combines macro-topic discovery (via LDA) with micro-semantic relationship analysis (via WCN). This integration directly addresses the semantic sparsity of injury reports, transforming flat keyword lists into interpretable causal networks, (2) theoretically, it extends the understanding of injury causation by utilizing network centrality metrics to identify and quantify the role of situational, implicit factors that act as critical mediators within specific injury pathways, offering a systemic perspective beyond frequency-based counts, and (3) practically, it provides a structured, network-based visualization of injury factors and their strongest associations, offering an evidence-based complement to key factors identification methods for prioritizing accident prevention measures.