Research on knowledge concept extraction method based on few-shot learning and chain-of-thought prompting

Knowledge concept extraction has important application value in the fields of education, medical care, and finance. Knowledge concept extraction is a sub-task of named entity recognition. However, due to the lack of data sets and the particularity of knowledge concept entity types, directly applying general named entity recognition methods to knowledge concept extraction tasks often has poor results. In view of the above challenges, a method based on few-shot learning and chain-of-thought prompting for knowledge concept extraction was proposed, utilizing open-source large language models. Firstly, text representations focusing on entity semantics were trained through contrastive learning, and the relevance of the retrieved few-shot examples was enhanced using the <italic>K</italic>-nearest neighbors algorithm. Secondly, a method utilizing chain-of-thought prompting was adopted to present the samples, with the aim of improving the reasoning ability of large language models in knowledge concept extraction. Experimental results on multiple datasets demonstrate that the few-shot learning and chain-of-thought prompting for knowledge concept extraction method, onthe whole, has shown results superior over existing methods.