融合TransD和边缘嵌入的知识图谱实体对齐

Knowledge Graph Entity Alignment of Fusing TransD and Edge Embedding

实体对齐用于在两个不同的知识图谱中发现在真实世界里表示同一对象的两个实体,而基于TransE模型的实体对齐方法在处理一对多、多对一、多对多等复杂关系时存在缺陷,即准确度不高的问题,不能精准推算出具有相同关系的实体,忽视了实体的多样性.针对上述问题,本文提出使用TransD模型处理一对一和多对一的实体关系,但缺点是忽略了关系的内在相关性,本文进一步使用边缘嵌入方法解决实体对齐的关系三元组,使用TransDCP-Align模型嵌入关系结构,利用三元组嵌入获得每一个实体的向量表示,并迭代更新每一个实体的向量表示完成实体对齐.在真实世界数据集的实验表明,本文提出的TransDCP-Align模型比传统表示学习模型有更好的实体对齐性能.

Entity alignment is used to represent two entities of the same object in the real world in two different knowledge graphs.However,the entity alignment method based on the TransE model has defects in dealing with one-to-many,many-to-one,many-to-many complex relations,and has low accuracy in dealing with complex relations.The inability to accurately extrapolate entities with the same relationship ignores the diversity of entities.In order to solve the above problems,the TransD model is proposed to deal with one-to-one and many-to-one entity relationships,but the disadvantage is that the intrinsic correlation of relationships is ignored.The edge embedding model is further used to solve the relational triples of entity alignment,the TransDCP-Align model is used to embed the relational structure,and the vector representation of each entity is obtained by triplet embedding.And iteratively update each entity vector to indicate completion of entity alignment.Experiments on real world data sets show that the proposed TransDCP-Align model has better entity alignment performance than the traditional representation learning model.