融合上下文的知识图谱补全方法  

Context fusion-based knowledge graph completion

作　　者：那宇嘉 谢珺 杨海洋 续欣莹 NA Yujia;XIE Jun;YANG Haiyang;XU Xinying(College of Information and Computer,Taiyuan University of Technology,Jinzhong 030600,Shanxi,China;College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,Shanxi,China)

机构地区：[1]太原理工大学信息与计算机学院,山西晋中030600 [2]太原理工大学电气与动力工程学院,山西太原030024

出　　处：《山东大学学报（理学版）》2023年第9期71-80,共10页Journal of Shandong University(Natural Science)

基　　金：山西省科技合作交流专项项目(202104041101030)。

摘　　要：构建了一种融合了上下文的知识图谱补全模型。首先通过Inception网络得到给定头尾实体对的深度交互嵌入;其次定义和编码给定实体对的2种上、下文信息:邻接关系上下文和路径上下文;然后使用基于头尾交互嵌入的注意力机制,分别聚合给定实体对的邻接关系上下文和路径上下文;最后利用全连接层来融合给定实体对的2种上下文信息,预测给定实体对之间的关系。在数据集FB15K-237、WN18RR和NELL-995中与其他主流模型对比,实验结果证实了所提补全模型的有效性。A knowledge graph completion model integrating context is constructed.Firstly,the deep interactive embedding of a given head-tail entity pair is obtained through the Inception network.Secondly,two types of context information of a given entity pair are defined and coded:adjacency and path context;Next,the attention mechanism based on head-to-tail interaction embedding is used to respectively aggregate the adjacency and path context of a given entity pair;Finally,the full connection layer is used to fuse the two types of context information of a given entity pair and consequently predict the relationship between the given entity pairs.Compared with other mainstream models in datasets FB15K-237,WN18RR and NELL-995,the experimental results show that the proposed model is effective.

关 键 词：知识图谱补全 Inception网络 邻接关系 关系路径 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]