一种结合实体邻居信息的知识表示模型  

Knowledge Representation Model That Combining Entity Neighbor Information

作　　者：洪锦堆 陈伟[1] 赵雷[1] HONG Jin-dui;CHEN Wei;ZHAO Lei(School of Computer Science and Technology,Soochow University,Suzhou 215006,China)

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006

出　　处：《小型微型计算机系统》2020年第8期1596-1601,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61572335)资助;江苏省高等院校自然科学研究重大项目(19KJA610002)资助。

摘　　要：引入图像、属性、实体描述文本等来自外部的信息有助于丰富知识表示模型中的实体向量表示.但是,外部信息并不总是有效而且目前的引入方法往往效率较低.针对以上问题,本文提出了一种结合实体邻居信息的知识表示模型,该模型把从知识图谱内部获取的实体邻居作为引入的信息,然后利用自动关键词抽取技术从实体邻居中选取出部分关键的邻居,最后使用本文提出的短接联合表示方法高效地将选出的邻居结合到知识表示模型中.实验结果表明,该模型在知识图谱的链接预测任务上优于目前的最优方法.Introducing external information such as images,attributes,entity description texts etc.into know ledge graph is helpful to enrich the representation of entity vectors in know ledge representation model.However,the external information is not always effective and the existing methods for introducing information are often inefficient.To solve the above problems,a know ledge representation model which combining entity neighbor information is proposed.The model first takes the entity neighbors obtained from the know ledge graph inside as the introduced information,and then uses the automatic keyword extraction technology to select some key neighbors from the entity neighbors.Finally,the short joint representation method proposed in this paper is used to efficiently combine the selected neighbors into the know ledge representation model.The experimental results show that the proposed models achieving the state-of-the-art in evaluating metrics on link prediction task.

关 键 词：知识表示学习 实体邻居 联合表示 链接预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]