Application of DA-Bi-SRU and Improved RoBERTa Model in Entity Relationship Extraction for High-Speed Train Bogie

Due to the large number of professional terms and complex entity relationships in the field of high-speed train (HST) bogie, the accuracy of entity relationship extraction is low. In order to improve the efficiency and accuracy of entity relationship extraction in high-speed train bogie domain, we propose a novel entity relationship extraction model for the domain of high-speed train (HST) bogie with the aim of improving the efficiency and accuracy of entity relationship extraction. The proposed model is based on RoBERTa-wwm (A Robustly Optimized BERT Pretraining Approach with Whole Word Masking) and DA-Bi-SRU (Double-Attention-Based Bidirectional Simple Recurrent Unit). To facilitate this, we construct a new bogie relation extraction dataset comprising of 25,000 statements collected from literature and professional annotations. The RoBERTa-wwm is employed to obtain dynamic word vectors from the input statements and optimized using the bogie dataset. Subsequently, a Bi-SRU model based on dual attention mechanism is developed to capture bidirectional semantic information and contextual semantic linkage in a rapid manner. Our experiments show that the RoBERTa-wwm-DA-Bi-SRU model outperforms Bi-LSTM and RNN methods with a prediction accuracy of 88.53% and an F1 value of 86.60%. Our proposed model thus demonstrates the potential to accurately extract entity relationships in the bogie knowledge graph of high-speed trains, simplifying the construction process.