Small sample fine-grained text named entity classification method based on prototype comparative learning

The invention belongs to the technical field of computers, and discloses a small-sample fine-grained text named entity classification method based on prototype comparative learning, which performs multi-granularity hierarchical prototype comparative learning by using a small number of labeled samples, and combines prototype-prototype comparative loss, prototype-instance comparative loss and instance-instance comparative loss to classify named entities of small-sample fine-grained text. And class prototype representation with inter-class hierarchical relationships is constructed, so that the ability of cognizing and summarizing things with hierarchical relationships is obtained, and accurate class-level semantics are learned from a small amount of data. 本发明属于计算机技术领域，公开了一种基于原型对比学习的小样本细粒度文本命名实体分类方法，通过利用少量标注样本，进行多粒度层次原型对比学习，结合原型-原型对比损失、原型-实例对比损失、实例-实例对比损失，构建具有类间层次关系的类原型表示，从而获得对具有层级关系的事物认知和概括的能力，从少量数据中学习准确的类级语义。