计算机应用研究2025，Vol.42Issue(6)：1632-1640,9.DOI:10.19734/j.issn.1001-3695.2024.11.0467
基于生成对抗网络与渐进式融合的多模态实体对齐
Multimodal entity alignment based on dual-generator shared-adversarial network
摘要
Abstract
In the field of education,knowledge graph fusion plays a crucial role.As a core technology of knowledge graph fu-sion,entity alignment aims to identify equivalent entity pairs across multiple knowledge graphs.Most existing entity alignment methods assume that each source entity has a corresponding entity in the target knowledge graph.However,when using cross-lingual and cross-graph entity sets,the problem of dangling entities arises.To address this issue,this paper proposed the dual-generator shared-adversarial network entity alignment model(DGSAN-EA).This model utilized partial parameter sharing and an optimal selection strategy to train two generators,selecting the optimal generator to conditionally generate new entities across knowledge graphs,thereby enhancing the dataset and solving the dangling entity problem.Furthermore,a progressive fusion strategy and the introduction of a distribution consistency loss function effectively resolve the distortion of fused feature informa-tion and the misalignment between modalities in multimodal entity alignment.Validation on multiple public datasets shows that compared to existing multimodal entity alignment models,DGSAN-EA achieves higher hit@k and MMR scores,demonstrating its effectiveness in entity alignment tasks.关键词
知识图谱/实体对齐/对抗网络/双生成器/参数共享/渐进式融合/分布一致性Key words
knowledge graph(KG)/entity alignment/adversarial network/dual generator/parameter sharing/progressive fusion/distribution consistency分类
信息技术与安全科学引用本文复制引用
冯广,郑润庭,刘天翔,杨燕茹,林健忠,钟婷,黄荣灿,项峰,李伟辰..基于生成对抗网络与渐进式融合的多模态实体对齐[J].计算机应用研究,2025,42(6):1632-1640,9.基金项目
国家自然科学基金重点项目(62237001) （62237001）
广东省哲学社会科学青年项目(GD23YJY08) （GD23YJY08）
