基于属性增强的对偶图实体对齐算法  

Attribute Augmentation Based Alignment Algorithm for Pairs of Dyadic Graph Entity Alignment

作　　者：姚荣 

机构地区：[1]成都信息工程大学软件工程学院,四川 成都

出　　处：《计算机科学与应用》2023年第5期1166-1177,共12页Computer Science and Application

摘　　要：本文提出结合属性信息的对偶图实体对齐算法针对基于关系感知的双对偶关系图算法中没有考虑到的属性信息进行优化,对属性结构嵌入向量使用图卷积神经网络算法对邻居节点抽取信息,并使用对偶关系图和注意力机制抽取实体对中的关系信息,最后通过结合实体对的关系信息和属性信息的相似度,判断是否为同一实体。针对原算法中识别效率不高的异构知识图谱实体对提升效果明显。在数据集DBP15K的三个跨语言数据集ZH-EN,JA-EN,FR-EN上实验,实验结果验证了对偶注意力以及属性信息对实体对齐方法的有效性。In this paper, we propose the dyadic graph entity alignment algorithm DAI combining attribute information to optimize the attribute information that is not considered in the dual dyadic relational graph algorithm RDGCN based on relationship awareness, use the graph convolutional neural network algorithm for attribute structure embedding vector to extract information from neighbor nodes, and use the dyadic relational graph and attention mechanism to extract the relational in-formation in entity pairs, and finally by combining the relational. The similarity of relationship information and attribute information of entity pairs is finally judged whether they are the same entity or not. The improvement effect is obvious for the heterogeneous knowledge mapping entity pairs which are not recognized efficiently in the original algorithm. Experiments on three cross-lingual datasets ZH-EN, JA-EN, and FR-EN of dataset DBP15K are conducted, and the experi-mental results verify the effectiveness of pairwise attention and attribute information on entity alignment methods.

关 键 词：属性增强 对偶关系图 实体对齐 知识图谱 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]