Aircraft accident investigation reports contain important information regarding the chronology of events, findings, contributing factors, and safety recommendations that can be used to understand accident patterns. However, this information is generally presented in the form of unstructured text, making manual analysis less efficient. This research aims to apply BERTopic to identify latent themes in aviation accident investigation reports published by the National Transportation Safety Committee (KNKT). A total of 182 investigation reports classified as Final Reports were processed through text extraction, corpus formation, and preprocessing, resulting in 172 documents used for topic modelling. BERTopic was implemented using sentence embedding, UMAP dimensionality reduction, HDBSCAN clustering, and c-TF-IDF-based topic representation. The modelling results yielded ten main topics reflecting various aspects of aviation safety, including technical, operational, and human factors. Evaluation showed that BERTopic achieved a topic coherence (Cv) value of 0.6111 and generated more specific keywords compared to Latent Dirichlet Allocation (LDA). The research results indicate that BERTopic is capable of effectively extracting latent themes from KNKT investigation reports and has the potential to support the analysis of aviation accident patterns and serve as a foundation for the development of domain knowledge-based research in the field of aviation safety.
Adelia Ramadhina Azzahra, K. M. Hindrayani, Andri Fauzan Adziima· ICEETE Conference Series· 0 citations
Retrieval-Augmented Generation (RAG) has improved the factual grounding of large language models; however, conventional retrieval strategies remain limited for long-document question answering, as relevant information is often distributed across multiple document sections and may be inferential rather than lexically similar to the query. This paper proposes a reasoning-aware hierarchical traversal mechanism that interleaves chain-of-thought generation at each depth-first search node evaluation step, incorporating the resulting reasoning embedding into a combined node scoring function alongside query similarity. The hierarchical tree is constructed through iterative chunking, embedding, clustering, and summarization; retrieval is then guided by this combined score under a dual-threshold pruning mechanism that adaptively controls traversal depth and breadth. The proposed approach is evaluated on the NarrativeQA and QuALITY benchmarks against a semantic similarity-based traversal baseline. On QuALITY, the method achieves marginal gains in overall accuracy (+0.7%) and F1 (+0.1%), while substantially reducing the abstain rate from 4.1% to 1.8%, with improvement concentrated on normal-difficulty questions; on hard questions, the baseline outperforms the proposed method. On NarrativeQA, ROUGE-L F1, BLEU-1, and BLEU-4 improve slightly, though METEOR decreases, reflecting inconsistent metric-level effects. These results suggest that step-wise CoT reasoning can improve node selection quality in hierarchical traversal, with gains most evident on standard-difficulty questions. Performance on hard questions and metric-level consistency across NarrativeQA remain open challenges, indicating that the quality of the generated reasoning signal is a key bottleneck for further improvement.
