Analysing aviation safety reports for uncovering current risks and policy gaps using topic modelling and knowledge graphs

• We leverage NLP and LLM to analyse aviation safety investigation reports. • Develop an integrated framework combining topic modelling and knowledge graph analysis. • Identify key safety concerns, including helicopter operations, weather factors, and engine failure. • Provide insight into current aviation safety issues in Australia. • Consider future challenges, including climate change and advanced air mobility. This paper analyses air transport safety investigation reports to reveal region-specific aviation safety hazards and identify current and potential future environmental, regulatory and operational risk factors. We present a novel framework that leverages advanced natural language processing (NLP) and large language model (LLM) techniques, using topic modelling and knowledge graphs for the analysis of said investigation safety reports. A key objective is to pinpoint safety issues in Australian aviation through deployment of an optimised BERTopic approach for topic modelling. An LLM is further used for entity extraction and knowledge graph construction to map relationships between safety factors. Our results reveal 22 distinct clusters, achieving a superior coherence score of 0.58 compared to conventional models, and by incorporating knowledge graph analysis, we better reveal the complex interdependencies between different safety factors. By mapping the outcomes of the proposed framework to well-known aviation safety models, we provide early signals of risks that may become critical in the future, including advanced air mobility and more extreme weather conditions. This positions our framework as diagnostic, enabling investigators to better understand safety factor interdependencies, identify risk concentrations, and enhance preparedness for evolving aviation operational contexts.