Named Entity Recognition for Automotive Fault Domain Based on Adaptive Transformer
Chenyi Li, Yongmao Zhao, Fei Tang · 2025
Aiming at the problems of complex hierarchical relationships between entities and lack of ability to perceive entity boundaries in the field of automotive equipment failure and repair, which are difficult to be effectively solved by the current mainstream models based on sequence annotation, an entity recognition method based on adaptive Transformer and lexical enhancement for the field of automotive failure is proposed. The method fuses Chinese character vectors and lexical information in the embedding layer, uses an improved Transformer model with direction and distance sensing ability in the coding layer to capture the long-distance-dependent information of the text sequences and the close relationship between entities, and finally decodes the output of the coding layer by conditional random field to obtain the optimal labeled sequences. The F1 value of this method on the constructed Chinese automotive equipment failure and repair domain dataset is improved by 2.91% compared with the baseline model, and the experimental results show that the fusion of lexical feature information can improve the model's ability to sense entity boundaries in the field of automotive failures.