A Named Entity Recognition Method in the Winter Olympics Domain Based on the Fusion of BERT and Domain-specific Static Word Embeddings

Aiming at the shortcomings of BERT in identifying domain-specific entity boundaries under limited data and computational resources, this paper explores an entity recognition model named BWBLC, which integrates BERT's dynamic contextual information with static word embeddings in the Winter Olympics domain and applies the fused vectors to the BiLSTM-CRF model for entity recognition. This fusion method can comprehensively utilize both global and local information to achieve more comprehensive information utilization, effectively compensating for BERT's shortcomings in identifying domain-specific entity boundaries under limited data and computational resources, thereby improving the performance of named entity recognition tasks in the Winter Olympics domain. The comparative experimental results on the dataset show that the proposed model outperforms other models in terms of macro-average precision, macro-average recall, and macro-average Fl score.