Graph-Enhanced Deep Learning With Character Similarity Mining for Automated NOTAM Correction in Aviation Systems

As the number of Notices to Airmen (NOTAMs) has increased exponentially, the existing manual processing methods have become increasingly inefficient and error-prone. To address this challenge, we used raw NOTAM data from the Civil Aviation Information Center, from September 2021 to September 2023. A trained relational graph convolutional neural network (CEV-RGCN) model, along with a tianzege-convolutional neural network (CNN), were employed to calculate the feature similarity of Chinese characters based on phonetics and glyphs. Based on this analysis, separate knowledge graphs for phonetics and glyphs were constructed, forming the foundation for a similar character library. During the masking stage of the CKBERT model, both positive and negative samples of knowledge graph triples were generated and integrated into multi-hop comparative learning. Candidate characters are retrieved from the similar character database, and the appropriate character is selected for replacement in the original text based on contextual information and similarity. This led to the development of the Ctc-CKBERT model, which significantly enhanced both computational efficiency and accuracy. Characters with a similarity score ranging from 0.9 to 1 were identified and applied based on the NOTAM dataset, thereby improving the accuracy of text error correction.