A LEBERT-Based Model for Named Entity Recognition
Shuang Li, Ziqiang Bao, Shuai Zhao, Guisong Jiang, Linlin Shan, Long Zhang · 2021 3rd International Conference on Artificial Intelligence and Advanced Manufacture · 2021
Recently, many works have tried to augment the performance of Chinese named entity recognition (NER) using word lexicons. Because traditional named entity recognition models cannot integrate lexical information into embeddings, the LEBERT-BiLSTM-CRF model is proposed for elementary mathematics text NER. Lexicon Enhanced BERT(LEBERT) integrates external lexicon knowledge into BERT layers directly by a lexicon adapter layer. We can get the embedding after training the LEBERT model. Then put it into BiLSTM for feature extraction. Finally, the CRF is used for correction. Experiments on datasets show that LEBERT-BiLSTM-CRF outperforms BiLSTM-CRF baselines, and F1 scores reached 95.02%. Compared with other NER models, the LEBERT-BiLSTM-CRF model also performs better.