Study of Bi-loop BERT-based Corner Tagging Joint Event Extraction

A Bi-loop BERT-based Corner Tagging (Bi-BCT) joint event extraction model is introduced to enhance aviation accident report analysis by addressing cascading errors and redundancy. Leveraging pre-training models, triggers are identified. Context and features are merged to compute argument confidence. Corner tagging matrix is used for triples decoding. Textual specifics of aviation accidents are analyzed to expand the Aviation Accident Report (AAR) dataset, boosting its quality by 7.05%. Experimental results on the AAR dataset reveal Bi-BCT outperforms the Onerel model with a 9.2% accuracy, 7.2% recall, and 8.2% F1 value improvement, showcasing its efficacy in aviation accident information extraction.