Named Entity Recognition: A Comparative Study of Advanced Pre-trained Model

Abstract

The increasing demand for healthcare has positioned digital medical services as a pivotal trend in the medical industry's future. This paper examines the role of Medical Named Entity Recognition (NER) in enhancing the utilization of medical resources and promoting the intelligentization of clinical decision-making. Medical NER faces challenges due to the diversity and specialized nature of medical text data, but with the application of deep learning technologies, there has been a significant improvement in recognition accuracy and efficiency. A novel model, RoBERTa-FGM-MHA-CRF, is proposed in this study, which integrates adversarial training, multi-head attention mechanisms, bidirectional Long Short-Term Memory (LSTM) networks, and bidirectional Gated Recurrent Unit (GRU) networks to substantially enhance the model's generalization and accuracy in medical NER tasks. Experimental results using a dataset of medicine instruction manuals demonstrate the model's effectiveness in practical medical text processing.