Named Entity Recognition Method with External Knowledge Injection in Low-resource Environments

Few-shot Named Entity Recognition requires models to rapidly learn the ability to recognize new entity types from a limited amount of training data. Previous research has focused on transferring knowledge from high-resource environments to low-resource ones but lacks relevant studies on the impact of knowledge contained in external knowledge graphs on model capabilities. We propose the KIner model, which, during the fine-tuning stage, injects external high-quality knowledge into the model using fixed templates, conveniently enhancing the model's adaptability to new entity types. Through experimental comparisons, our model achieves promising results in six different benchmark tests.