Conceptual

Comparative Topic Modeling of Aviation Safety Narratives with NLP

An empirical comparison of four topic-modelling techniques - pLSA, LSA, LDA, and Non-negative Matrix Factorization - applied to free-text aviation incident narratives from the ATSB dataset. After NLP preprocessing and building a document-term matrix, each method extracts ten latent safety themes (bird strikes, landing gear, air traffic control, engine issues, and more), and the paper contrasts their interpretability, scalability, and sensitivity to preprocessing and hyperparameters to guide analysts choosing a method for large incident-report corpora.