Chinese Nested Named Entity Recognition in the Medical Field Based on BiLSTM and Multi-Head Global Pointer

The technology of named entity recognition has been gradually refined. However, current research is primarily focused on non-nested named entity recognition, with less attention devoted to nested named entity recognition (NNER). The challenge of accurately extracting nested entities from complex medical texts remains. To address the issue of low accuracy in identifying nested entities in medical named entity recognition, we propose a Chinese nested named entity recognition model (Multi-ABGP) based on adversarial training, bidirectional long short-term memory (BiLSTM) and multi-headed global pointer network. This model employs a BiLSTM network to extract bidirectional and long-distance text features, enhancing the ability of the global pointer network to extract features from longer text. Furthermore, adversarial training is employed to reduce the impact of noise on the model's performance and enhance its robustness. This enables the model to ignore irrelevant features introduced by noise and instead focus on entity-related features. Finally, a multi-head global pointer network is utilized to fuse the features of multiple global pointer networks. This reduces the erroneous prediction of a single pointer network and improves the overall performance of the model. The experimental results demonstrate that the proposed model achieves Micro-F1, Micro-P, and Micro-R values of 73.71%, 74.96%, and 72.50%, respectively, on the CMeEE-V2 dataset of Chinese nested named entity recognition in the medical specialty. This significantly improves the effectiveness of Chinese nested entity recognition in the medical field. And we also compare it with the large language model (LLM), the experimental results show that the model outperforms the ChatGLM in the nested entity recognition task. Additionally, it exhibits favorable results on the non-nested entity recognition dataset People's Daily.