命名实体消歧研究综述

Review of Named Entity Disambiguation Studies

实体消歧是指在一个具体的知识库中,把一个被标识的实体指称链向它对应条目的过程。实体消歧的任务是根据上下文信息解决一个命名实体指称项对应多个实体概念的一词多义问题,它在从海量数据准确提取信息的知识图谱构建过程中起到重要作用,是自然语言处理中的一项基本任务。该文主要对实体消歧技术的相关研究内容进行综述。首先,阐述了实体消歧的国内外研究背景,并对命名实体识别、候选实体生成、候选实体排序等实体消歧相关理论进行全面梳理。其次,对实体消歧的具体含义及其研究内容进行详细综述,并对实体消歧研究内容的特点进行了分析。再次,将实体消歧技术的实现方法划分为三类并对涉及到的数据集进行归纳,并从四个方面讨论了实体消歧领域存在的难点和提高实体消歧准确率的途径,对消歧方法的优缺点及评价指标进行了总结,意在为改善实体消歧效果提供新的解决思路。最后,对实体消歧技术的应用和发展前景进行总结。

Entity disambiguation is the process of chaining an identified entity referent to its corresponding entry in a specific knowledge base.The task of entity disambiguation is to solve the word polysemy problem where a named entity referent term corresponds to multiple entity concepts based on contextual information,and it plays an important role in the construction of knowledge graphs for accurate extraction of information from massive data,which is a fundamental task in natural language processing.We mainly review the research content related to entity disambiguation techniques.Firstly,the background of the domestic and international research on entity disambiguation is described,and the theories related to entity disambiguation such as named entity identification,candidate entity generation,and candidate entity ranking are comprehensively reviewed.Secondly,a detailed overview of the specific meaning of entity disambiguation and its research content is presented,and the characteristics of the research content of entity disambiguation are analyzed.Thirdly,the implementation methods of entity disambiguation techniques are classified into three categories and the data sets involved are summarized,and the difficulties in the field of entity disambiguation and the ways to improve the accuracy of entity disambiguation are discussed from four aspects,and the advantages and disadvantages of disambiguation methods and evaluation indexes are summarized,with the intention of providing new solutions for improving the effectiveness of entity disambiguation.Finally,the application and development prospects of entity disambiguation techniques are summarized.