A Low-Resource Chinese Named Entity Recognition Method Combining Chinese Glyph Features and Contrastive Learning

Abstract Named Entity Recognition (NER) represents a pivotal research area in the domain of natural language processing, yet the effective utilization of Chinese information remains a significant challenge. Moreover, NER tasks often suffer from limited data availability, data with varying labeling quality, and potential ethical concerns. To address these challenges, we propose a novel approach for low-resource Chinese named entity recognition by leveraging Chinese glyph features and contrastive learning. Our method effectively enhances the accuracy of named entity recognition. Through extensive experimentation, we demonstrate the efficacy of our approach on both the low-resource medical dataset for esophageal cancer and general Chinese dataset. Our model outperforms the widely adopted BERT-CRF model on the medical dataset, achieving a precision improvement of 3.12\%, a recall improvement of 1.73\%, and an F1 score improvement of 2.62\%. Notably, our core contrastive learning framework can be applied not only to the BERT model but also to the majority of Chinese NER task models, exhibiting its versatility and potential impact on the broader NER research field.