Patterns Associated with Fatal Motorcycle-Involved Crashes in Bangladesh: Applying Text Mining Techniques and Structural Topic Modeling

Motorcycle crashes in Bangladesh pose a growing public health and transportation safety challenge, worsened by persistent gaps and availability in official crash data. This study uses a news media mining approach to examine 475 motorcycle-related crash reports from English-language Bangladeshi online newspapers published in 2022, forming a corpus of 50,190 words. Combining text mining with Structural Topic Modeling (STM) approach, the study identifies key crash patterns and latent themes within unstructured narratives. Preliminary analyses reveal frequent references to severe injuries, emergency response, young riders, fixed-object collisions, and nighttime crashes. Rapid Automatic Keyword Extraction and word co-occurrence network analyses underscore high-risk behaviors such as speeding and distraction, vulnerability of young male riders and students, and the influence of rural roads, highways, and urban environments. STM results indicate two dominant crash profiles: fatal single-vehicle crashes in rural settings linked to loss of control, and multiple-vehicle crashes in urban or highway areas often involving heavy vehicles. Guided by the Safe System Approach, the study proposes context-specific countermeasures, including infrastructure improvements, lane discipline enforcement, youth-focused education, improved vehicle visibility, and stronger emergency medical services. Overall, the findings demonstrate that online news media provide a scalable and informative source for motorcycle safety research in data-limited settings, supporting targeted strategies.