ERNIE-AT-CEL: A Chinese Few-Shot Emerging Entity Linking Model Based on ERNIE and Adversarial Training.

This article proposes a Chinese few-shot emerging entity linking model based on ERNIE and adversarial training. The model utilizes ERNIE as the base model and achieves accurate linking of Chinese few-shot emerging entities by adding adversarial perturbations during the training process. Experimental results on standard entity linking evaluation datasets demonstrate significant performance improvements of our proposed model compared to baseline models. Moreover, we compare multiple candidate entity retrieval methods through comparative experiments to evaluate and compare their effectiveness in the entity linking task. The experimental results show that our appoarch achieved an F1 score of 0.60 and ranked second in the NLPCC 2023 Shared Task 6 (Chinese Few-shot and Zero-shot Entity Linking). Experimental results demonstrate that the model has high predictive performance and robustness in the Chinese few-shot emerging entity linking task, providing reference and inspiration for research and practice in related fields.