面向小样本命名实体识别的标签语义增强原型网络

Enhanced Prototype Network with Label Semantics for Few-Shot Named Entity Recognition

小样本命名实体识别任务旨在通过有限数量的标注样本来识别并分类文本中的实体。目前,基于两阶段的小样本命名实体识别方法存在泛化能力差和原型类别混淆等问题。为了解决这些问题,该文提出了一种利用标签语义增强实体表示的两阶段方法。具体来说,该文使用蕴含语义信息的标签名称增强实体表示,并将其应用于跨度检测和实体分类模型。在跨度检测模型中,采用注意力机制将标签语义融入到文本表示,以减少跨度检测模型泛化能力不足的问题。同时,利用增强后的实体表示构建类别原型,使得原型可以获得更丰富的特征,从而降低了原型间的混淆。实验结果表明,该文方法可以充分利用标签语义信息,并在多个基准数据集上取得了良好的性能表现。

Few-shot named entity recognition aims to identify entity types with limited annotated data.Currently,the prototype network is usually employed for entity recognition,which is defected in distinguishing new category data.To address issue,this paper proposes a two-stage few-shot named entity recognition method enhanced by label semantics.Specifically,this paper uses the attention mechanism to incorporate label semantic information into the span detection.At the same time,this label semantic information also enhance specific types of span prototypes,making prototypes capture richer features and better distinguish different types of entities.Experimental results on multiple benchmark datasets show that the proposed method can achieve good performance.