A Multi-Task Instruction with Chain of Thought Prompting Generative Framework for Few-Shot Named Entity Recognition

Few-shot Named Entity Recognition (NER) is the task of identifying new named entities using only a small number of labeled examples. Prompt-based learning has been successful in few-shot NER by using prompts to guide the labeling process and increase efficiency. However, previous prompt-based methods for few-shot NER have limitations such as high computational complexity and insufficient few-shot capability. To address these concerns, we propose a multi-task instruction framework called CotNER for Few-shot NER, which utilizes a chain-of-thought prompting generative approach. We introduce two auxiliary tasks, entity extraction and entity recognition, and integrate reasoning processes through chain-of-thought prompting. Our approach outperforms previous methods on various benchmarks, as demonstrated by extensive experiments.