Chinese Named Entity Recognition Based on Lexicon Information Enhancement and Adaptive Adversarial Training

Aiming at the problem of insufficient accuracy enhancement of fused lexicon information in Chinese named entity recognition, this paper proposes a Chinese named entity recognition based on lexical information enhancement and adaptive adversarial training. The model directly integrates external lexicon knowledge into the BERT layer through the lexicon adapter. In addition, we also propose an adversarial training method. During the training process, this method dynamically adds perturbations to the network variables, aiming to enhance the robustness and generalization ability of the model. The model has experimented on two benchmark Chinese NER datasets, Resume and Weibo, and the experimental results show the effectiveness of the model.