RBIA_CNER: A Named Entity Recognition Method for Ethnic Educational Resources based on RoBERTa-wwm-ext Optimization

Aiming at the problems of insufficient data annotation and complex entity types in NER of ethnic education resources, a new NER method, RBIA_CNER, is proposed in this paper. In this method, RoBERTa-wwm-ext is used as the pre-trained language model, and the BiLSTM-IDCNN model is combined for feature extraction, effectively fusing global and local information, allowing the method to better understand the complex structure and details in the text. The output of BiLSTM processing is regularized by the Dropout layer, enhancing the generalization ability of the model. After the feature extraction layer, a 12-head multi-head attention mechanism is added to assign weights to the extracted features, weakening irrelevant features and enhancing key features. Finally, a linear layer is employed to map the features into the label space. The global optimal label sequence is generated using CRF decoding, which enhances the accuracy and consistency of entity recognition. Experiments were carried out on the dataset of ethnic education resources and public datasets. Experimental results show that the RBIA_CNER model achieves improvements of 3.0% in precision, 1.4% in recall, and 2.6% in F1 score compared to the baseline model BERT_BiLSTM_CRF on the ethnic education resources dataset. The results from public datasets demonstrate that RBIA_CNER surpasses the baseline model across all three metrics. This underscores the effectiveness of the proposed NER method for ethnic education resources in enhancing both recognition accuracy and overall performance.