Chinese Biomedical NER Based on Self-attention and Word-Relation Decoding Strategy

Biomedical named entity recognition plays a crucial role in advancing smart healthcare tasks. However, the scarcity of biomedical data and the extensive annotation required by professionals make achieving remarkable model performance challenging and expensive. Few-shot learning focuses on improving the model’s performance and generalization under limited labeled data, providing effective solutions for biomedical information mining. Therefore, this paper proposes a Chinese biomedical named entity recognition method based on self-attention and word-relation decoding strategy. The aim is to effectively address the task of Chinese biomedical named entity recognition in few-shot scenarios. Our work is based on the 9th China Health Information Processing Conference task 2 and ranked third among all the teams. In the final results of the query set in the test dataset, the F1 score on testA dataset is 0.85, and on testB dataset, it is 0.87.