Hierarchical Multilabel Classification for Fine-Level Event Extraction from Aviation Accident Reports

Large numbers of accident reports are recorded in the aviation domain, which greatly values improving aviation safety. To better use those reports, we must understand the most important events or impact factors according to the accident reports. However, the increasing number of accident reports requires large efforts from domain experts to label those reports. To make the labeling process more efficient, many researchers have started developing algorithms to automatically identify the underlying events from accident reports. This article argues that we can identify the events more accurately by leveraging the event taxonomy. More specifically, we consider the problem to be a hierarchical classification task, where we first identify the coarse-level information and then predict the fine-level information. We achieve this hierarchical classification process by incorporating a novel hierarchical attention module into the bidirectional encoder representations from transformers model. To further utilize the information from event taxonomy, we regularize the proposed model according to the relationship and distribution among labels. The effectiveness of our framework is evaluated using data collected by the National Transportation Safety Board. It has been shown that fine-level prediction accuracy is highly improved and that the regularization term can be beneficial to the rare event identification problem.