命名实体识别技术综述

Overview of Named Entity Recognition Technology

命名实体识别是自然语言处理中的热点研究方向之一,目的是识别文本中的命名实体并将其归纳到相应的实体类型中。首先阐述了命名实体识别任务的定义、目标和意义,分析提出了命名实体识别的主要难点在于领域命名实体识别局限性、命名实体表述多样性和歧义性、命名实体的复杂性和开放性;然后介绍了命名实体识别研究的发展进程,从最初的规则和字典方法到传统的统计学习方法再到现在的深度学习方法,不断地将新技术应用到命名实体识别研究中以提高性能;接着系统梳理了当下命名实体识别任务中的若干热门研究点,分别是匮乏资源下的命名实体识别、细粒度命名实体识别、嵌套命名实体识别以及命名实体链接;最后针对评判命名实体识别模型的好坏,总结了常用的若干数据集和实验测评指标,并给出了未来的研究建议。

Named entity recognition is one of the valuable research directions in natural language processing.The purpose is to identify named entities from text and classify them into corresponding entity types.Firstly,this paper explains the definition,goals,and meaning of named entity recognition tasks,and analyzes the main difficulties of named entity recognition,namely,the limitations of domain named entity recognition,the diversity and ambiguity of named entity expressions,the complexity and openness of named entities.Then,it introduces the development roadmap of named entity recognition tasks,originally from rules and dictionary then traditional statistical learning methods,and deep learning methods currently,continuously improving the performance.Next,we systematically organized several popular research points,including low resources named entity recognition,fine-grained named entity recognition,nested named entity recognition,and named entity linking.And several commonly used datasets and experimental evaluation indicators for named entity recognition tasks are summarized,and future research recommendations are given.Finally,in order to estimate the quality of named entity recognition models,several commonly used datasets and experimental evaluation indicators are summarized,followed by our future research opinion.