Domain Knowledge Enhanced BERT for Chinese Named Entity Recognition

Digitalization of educational resources and revolutionizing knowledge frameworks are essential in achieving smart education. Knowledge graphs play a pivotal role in addressing knowledge representation, correlation, and sharing within digital education. Named Entity Recognition (NER) is a fundamental task in constructing knowledge graphs. This study introduces a Domain Knowledge-Enhanced BERT Chinese NER model, DK-BERT-CRF (Domain Knowledge BERT CRF), to address the deficiency of lexical information features within the Chinese NER task using the BERT pre-trained model. To construct an automated labeling dataset, we perform an automated labeling dataset construction based on ChatGPT, focusing on the example of computer science's data structures. We conduct experiments and evaluations using this dataset and the general CLUENER2020 dataset. Comparative experiments with BERT+CRF and BiLSTM+CRF are also conducted. The experimental results demonstrate that the DK-BERT-CRF model, enriched with domain knowledge, exhibits an improved F1 score compared to the other two models. Particularly, the DK-BERT-CRF model showcases enhanced F1 scores on the computer science data structure dataset after the incorporation of domain knowledge.