结合信息交互的人物实体链接

Person Entity Linking Combined with Information Interaction

实体链接是将文本中的实体提及链接到知识图谱中实体节点的任务,是自然语言处理许多下游任务的重要基础.而在各类实体中,人物实体承载了知识图谱中主要的事实组成部分,但由于存在大量重名导致人物链接难度大大增加.人物实体链接是人物知识图谱构建的重要一环,其目的是把一段文本所描述的人物实体链接到图谱中正确的实体节点上.由于目前中文人物实体链接数据集比较缺乏,而通用实体链接数据集大多覆盖多种类型实体并且规模比较有限,因此本文基于百科网页数据构建了新的大规模中文人物实体链接数据集SummaryEL和TextEL,并通过采样验证了数据集的质量.基于新构建的数据集,本文提出基于描述文本和实体属性信息交互的人物实体链接模型,有效地建立描述文本和知识图谱节点之间的联系.实验结果表明,本文所提出的人物实体链接模型取得较高的准确率,在SummaryEL和TextEL测试集上的平均准确率分别达到89.27%和87.43%.该模型可作为该任务未来研究工作的基准方法.新构建的数据集和实验代码将公开在github上.

Entity linking,the task of linking entity mentions in text to entity nodes in the knowledge graph,is an important foundation for many downstream tasks in natural language processing.And among the various types of entities,person entities carry the main factual components of the knowledge graph,but the presence of a large number of renames makes person linking much more difficult.Person entity linking is an important part of the construction of the person knowledge graph,the purpose of which is to link the person entities described by a piece of text to the correct entity nodes in the graph.Since there is a lack of Chinese person entity linking datasets,and most of the generic entity linking datasets cover multiple types of entities and are limited in size,this paper constructs new large-scale Chinese person entity linking datasets,SummaryEL and TextEL,based on wikipedia web data and verifies the quality of the datasets by sampling.We further propose a new entity linking model based on the interaction between texts and entity attributions information to effectively build the connection of the texts to the nodes of knowledge graph.The experimental results show that the proposed model achieves high accuracy rates,with average accuracy rates of 89.27%and 87.43%on the SummaryEL and TextEL test sets,respectively.The model can be used as a benchmark method for further work.The newly constructed dataset and experimental code will be publicly available on github.