Chinese Named Entity Recognition of Military Weapons Based on TaCL-BiLSTM-CRF
Kun Guo, Xinfeng Shu · 2023
To improve the recognition accuracy of Chinese named entities of the military weapons, a TaCL-BiLSTM-CRF based named entity recognition method is proposed, which aims to recognize weapon names, attributes, component structures and uses. Firstly, the token-aware contrastive learning (TaCL) pre-training method is used to extract text features, then the bidirectional long short-term memory (BiLSTM) algorithm is used to obtain contextual information, and finally, the conditional random field (CRF) is used to recognize entities. Experiments on Chinese military dataset show that, compared with the avalible methods, the method proposed has a better performance in precision, recall, F1-value, and effectively improves the recognition ability of Chinese military named entities.