基于虚拟邻域聚合和属性嵌入的实体对齐方法  

Virtual neighborhood aggregation and attribute embedding for entity alignment

作　　者：冯晓慧 张晓滨[1] FENG Xiao-hui;ZHANG Xiao-bin(School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048

出　　处：《计算机工程与设计》2025年第3期698-704,共7页Computer Engineering and Design

基　　金：陕西省自然科学基础研究计划基金项目(2023-JC-YB-568)。

摘　　要：为增强对新实体的泛化能力,提高实体对齐的效果,构建一种基于多跳虚拟邻域聚合和属性嵌入的实体对齐模型。利用随机游走的方式构建虚拟邻域,使用门控单元对中心实体的虚拟邻域信息进行聚合;对于属性三元组生成抽象属性三元组,通过预测指定属性的相关属性进行属性嵌入;引入关系损失,通过排序得到实体对齐的结果。在公开数据集上的实验结果表明,该模型的Hits@1提升0.1~0.3、Hits@10提升0.04~0.2、MRR提升0.02~0.3,验证了模型的有效性。To enhance the generalization ability to new entities and improve the entity alignment effect,an entity alignment model based on multi-hop virtual neighborhood aggregation and attribute embedding was constructed.Virtual neighbors were constructed by utilizing random walks.Gated units were used to aggregate virtual neighborhood information of the central entity.Abstract attribute triples were generated for attribute triples,and attribute embedding was conducted by predicting related attributes of the given attribute.The relation loss was introduced,and entity alignment results were obtained through sorting.Results of experiments on public datasets show that the model improves Hits@1 by 0.1-0.3,Hits@10 by 0.04-0.2,and MRR by 0.02-0.3,which validates the effectiveness of the model.

关 键 词：知识图谱 实体对齐 虚拟邻居 邻域聚合 属性嵌入 知识图谱融合 知识表示学习 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]