文献类型：期刊文献

中文题名：基于多方嵌入的逐步实体对齐方法

英文题名：Stepwise entity alignment method based on multi?party embedding

作者：刘雪丽[1];李燕[1];李春雨[1];刘悦悦[1]

第一作者：刘雪丽

机构：[1]甘肃中医药大学信息工程学院,甘肃兰州730000

第一机构：甘肃中医药大学信息工程学院（教育技术中心）

年份：2024

卷号：47

期号：13

起止页码：138

中文期刊名：现代电子技术

外文期刊名：Modern Electronics Technique

收录：CSTPCD;;北大核心:【北大核心2023】；

基金：中国高校产学研创新基金(2021LDA09002);甘肃省研究生创新创业基金(2023CXZX-770)。

语种：中文

中文关键词：知识图谱;实体对齐;多方嵌入;逐步实体对齐算法;互相选择;最小相似度阈值;知识图谱嵌入

外文关键词：knowledge graph;entity alignment;multi-party embedding;StepEA algorithm;mutual selection;minimum similarity threshold;knowledge graph embedding

摘要：大多数实体对齐方法对知识图谱信息的利用不够充分,没有考虑实体间的互相选择,忽略了现实生活中很多实体在对方知识图谱中不存在等价实体的事实。针对以上问题,提出一种基于多方嵌入的逐步实体对齐方法。该方法对三元组信息、邻域信息、实体名称的语义信息和字符串信息进行多方嵌入生成相似度矩阵,再通过所提出的逐步实体对齐算法将目前彼此最为相似且相似度大于最小相似度阈值的两个实体进行匹配,直到剩余所有实体的相似度都不大于最小相似度阈值时停止匹配,在确保等价实体准确匹配的前提下,减小不存在等价实体时发生错误匹配的概率。在DBP15K数据集上进行了三项实验,结果证明了该方法和逐步实体对齐算法的有效性,以及多方嵌入中每个模块的必要性。
Most entity alignment methods fail to fully utilize knowledge graph information,fail to consider mutual selection between entities,and ignore the fact that many entities in real life do not have equivalent entities in each other′s knowledge graphs.To address these issues,a stepwise entity alignment(StepEA)method based on multi-party embedding is proposed.This method embeds triplet information,neighborhood information,semantic information of entity names,and string information into a similarity matrix by multi-party embedding.Then,with the proposed StepEA algorithm,the two entities that are currently most similar to each other and have a similarity greater than the minimum similarity threshold are matched until the similarity of all remaining entities does not exceed the minimum similarity threshold.On the premise of ensuring accurate matching of equivalent entities,the probability of incorrect matching when there are no equivalent entities is reduced.Three experiments were conducted on the DBP15K dataset.The results show the effectiveness of this method and the StepEA algorithm,as well as the necessity of each module in multi-party embedding.