Chinese named entity recognition using adversarial training and attention mechanism

Abstract:
Chinese NER plays an important role in natural language processing as a fundamental task. The current character-based training approach for Chinese NER suffers from the inability to incorporate overall sentence features or character neighbourhood features into the character vector. In this paper, we designed a feature enhancement module with attention mechanism and BI-LSTM networks to enhance the feature extraction capability of the model and proposed the introduction of adversarial training as a method of enhancing data to improve the robustness of the model. It has been proven through extensive experiments that the accuracy of the model in Chinese NER task has been improved after the introduction of the adversarial training and feature enhancement modules. In our experiments, we choose the BERT-CRF network as our baseline model.