面向多语义关系的知识图谱表示学习方法  

Knowledge map representation learning method for multi semantic relations

作　　者：周航 刘学军 张伯君 ZHOU Hang;LIU Xue-jun;ZHANG Bo-jun(School of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China;Product Inspection Center,Nanjing Boiler and Pressure Vessel Inspection Institute,Nanjing 210028,China)

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816 [2]南京锅炉压力容器检验研究院产品监检中心,江苏南京210028

出　　处：《计算机工程与设计》2021年第1期220-225,共6页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2018YFC0808500、2017YFC0805605);江苏省重点研发计划基金项目(BE2017617)。

摘　　要：考虑到关系的多语义性以及不同实体和关系之间的确定性,提出一种面向多语义关系的知识图谱表示方法TransC。将关系划分为多条语义,构建关系的高斯混合模型;构建对应的云模型,获取最能表达该关系的语言值和确定性;将确定性作为权重,以加权欧式距离作为新的评分函数;使用多个真实的基准数据集对链接预测和三元组分类进行广泛的实验。实验结果表明,相较于现有的模型和方法,TransC在各项指标上都显示出其优越性。Considering the multi-semantic nature of relationships and the certainty between different entities and relationships,a multi-semantic relationship oriented knowledge graph representation method TransC was proposed.The relationship was divided into multiple semantics to construct a Gaussian mixture model of the relationship.A corresponding cloud model was constructed to obtain the language value and certainty that could best express the relationship.The certainty was used as the weight,and the weighted Euclidean distance was used as the new scoring function.Extensive experiments were performed on link prediction and triple classification using multiple real benchmark data sets.Experimental results show that,compared with the existing models and methods,TransC shows its superiority on various indicators.

关 键 词：数据描述 多语义 高斯分布 确定性 云模型 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]