<p>Named entity recognition (NER) refers to recognize entity spans from text and categorize them into predefined entity types, which is significant for tasks such as knowledge graph construction. However, most existing entity recognition methods perform poorly in professional domains, such as cybersecurity, since they rely heavily on large amounts of high-quality labeled data for model training. In this paper, we introduce prompt-based learning and reformulate the NER task into a generation problem, and propose a knowledge enhanced prompting framework for few-shot NER, which can identify the named entities based on few labeled data. Firstly, we present a knowledge-enhanced template generation method, which integrates the entity domain knowledge and uses BART to generate templates automatically, which helps to overcome the difficulties caused by manual template engineering. Secondly, we construct prompts for each entity type and feed them into the masked language model (MLM) for entity prediction, which do not need to enumerate all the candidate spans, thus reducing the computational complexity. Thirdly, an interaction detection module is designed for entity prediction to constrain the identified entity spans in the input text, which can reduce the uncertainty and unstructured problem for NER. Finally, we perform comparison experiments with existing competitive models on three public datasets and a cybersecurity dataset constructed for professional domain, and the experimental results demonstrate that our method outperforms the existing models on both rich-resource and few-shot settings.</p>