Using lda2vec Topic Modeling to Identify Latent Topics in Aviation Safety Reports

Abstract

The study of aviation safety report in the aviation industry usually relies on manually labeled data sets, and then classifies and models related problems, which have become insufficient in the face of increasingly rapid report data. Therefore, in this paper, we propose an unsupervised machine learning algorithm architecture to explore topics (rather than categories) for optimizing aviation safety report text mining research, and further conducts preliminary experiments with real data from the Aviation Safety Report System (ASRS). The experimental results show that compared with traditional classification modeling, this scheme can identify latent topics with higher interpretability, and the training data does not rely on manual labeling, which can greatly improve the efficiency and pertinence of aviation safety report research.

Authors

• Yonghui Luo
• Hongwei Shi

Keywords

• Computer Science
• Engineering