MKE: 基于背景知识与多头选择的嵌套命名实体识别  

MKE:Nested NER Based on Knowledge Embedding and Multi-Head Selection

作　　者：李政 涂刚[1] 汪汉生[2] LI Zheng;TU Gang;WANG Hansheng(School of Computer Science and Technology,Huazhong University of Science and Technology,Wuhan,Hubei 430074,China;China Ship Development and Design Center,Wuhan,Hubei 430064,China)

机构地区：[1]华中科技大学计算机科学与技术学院,湖北武汉430074 [2]中国舰船研究设计中心,湖北武汉430064

出　　处：《中文信息学报》2024年第4期86-98,107,共14页Journal of Chinese Information Processing

基　　金：国防基础科研计划(JCKY2019204A007)。

摘　　要：目前,在嵌套命名实体识别研究中,基于片段的方法将命名实体识别转化为分类问题,通过微调预训练模型,能够较好地识别嵌套实体,但仍存在领域知识缺乏和无法实现实体多分类的不足。该文提出基于知识嵌入的多头模型,用于解决这些问题。模型的改进包括:①引入领域背景知识,知识嵌入层以实体矩阵的形式,实现背景知识的无损嵌入;②将命名实体识别过程转化为多头选择过程,借助注意力打分模型,计算候选片段得分,最终在正确识别嵌套实体边界的同时实现实体多分类。实验结果表明,以实体矩阵方式实现的背景知识嵌入,可以有效提高识别准确率,在7个嵌套与非嵌套命名实体识别数据集上取得SOTA表现。In existing research on nested named entity recognition,this task is treated as span classification tasks via finetuned pretrained models.This paper proposes a multi-head model based on knowledge embedding(MKE for short)method to further improve this task.This method introduces domain-specific knowledge in the form of entity matrices,allowing the background knowledge to be embedded without any loss.It also transforms the named entity recognition into a multi-head selection process,followed by scoring the candidate spans using the attention score model.The experimental results show that the proposed method achieves the state-of-the-art performance on seven nested and flat named entity recognition datasets.

关 键 词：嵌套命名实体识别 知识嵌入 多头选择 注意力 实体多分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]