Abstract: The process text relationship extraction task is of strong application value. To deal with the numerous and complex relationships in process texts, this paper proposes the nearest neighbor prompt learning model K-PTR. Firstly, the model sets a label set for each relationship class in the prompt learning stage and summarizes the entity types and relationship contents of the relationship in the process text, which helps the model effectively improve the sample utilization rate in the case of few-shot. Then the KNN method is used to combine the similarity of instances to complete the reasoning and prediction of the relationship categories of the instances to be tested Experiments indicate that the F₁ values of K-PTR are increased by 1.2%, 1.9%, 1.8% and 2.5% in four -shot experiments, respectively.