Prompt-Based Data Augmentation Framework for Few-Shot Named Entity Recognition

Named Entity Recognition (NER) refers to the recognition of entities with specific meanings in natural language texts. In practical situations, NER is often limited by low-resource scenarios. NER suffers from an insufficient number of labeled samples in the target domain and the emergence of new entity types. To address these issues, this paper proposes a novel method for NER, utilizing prompts to identify entities in various domain texts under low-resource conditions. Specifically, we employ the following two methods: (1) Using a formalized data augmentation method to enhance data diversity; (2) During training, simulating conditions from the testing process as much as possible and introducing unknown information into the data. Our method effectively enhances its ability to handle unknown data, improving the model's generalization and adaptability. Experimental results demonstrate competitive performance across all datasets, with a 2.99% increase in the F1 score on the WNUT dataset.