太原理工大学学报2025，Vol.56Issue(5)：875-886,12.DOI:10.16355/j.tyut.1007-9432.20250249
基于判别性多特征融合的实体对齐算法
Entity Alignment Algorithm Based on Discriminant Multi-feature Fusion
摘要
Abstract
[Purposes]Knowledge Representation Learning(KRL)has achieved remarkable suc-cess in cross-language entity alignment.However,it fails to model the complex semantic relation-ships between heterogeneous knowledge graphs,and most existing methods rely on local feature matching,thus failing to make full use of the structural information of knowledge graphs.Therefore,in this paper,an Entity Alignment algorithm based on Discriminant Multi-feature Fusion(EA-DMF)is propsed.[Methods]This method utilized the semantic information,structural information,and at-tribute information in the knowledge graphs for multi-feature fusion to fully explore the potential se-mantic information in the graphs.Specifically,with EA-DMF,the Gromov-Wasserstein distance measure was introduced to measure the similarity between graphs,a random relation walk algorithm was established,the semantic information of entities was enriched by using the long-term dependency relationships in the knowledge graph,and the high-confidence local alignment information was gradu-ally extended to the global level through the iterative update of high-confidence anchor nodes.Finally,the multi-perspective optimal transmission theory was applied to fuse multiple information features and thereby the set of aligned entity pairs was abtained.[Results]After extensive experiments on five entity alignment datasets,EA-DMF surpasses multiple competing baselines without any supervision or hyperparameter adjustment,proving that this method can align unknown entities in knowledge graph more effectively and accurately.关键词
知识图谱/实体对齐/随机关系游走/渐进式优化/最优传输Key words
knowledge graph/entity alignment/random relation walk/progressive optimization/optimal transmission分类
信息技术与安全科学引用本文复制引用
申卫杰,王莉..基于判别性多特征融合的实体对齐算法[J].太原理工大学学报,2025,56(5):875-886,12.基金项目
国家自然科学基金资助项目(2021YFB3300503) （2021YFB3300503）
