融合实体邻域信息的时序知识图谱实体对齐

打开文本图片集

关键词：时序知识图谱；实体对齐；知识嵌入；实体活跃度；潜在邻居 中图分类号：TP391.1 文献标志码：A 文章编号：1001-3695（2025）07-017-2048-08 doi：10.19734/j. issn.1001-3695.2024.12.0513

Abstract：Thepurposeof entityalignment is tobuildacomprehensive knowledgegraphbymatching identical entitiesacross multi-source knowledge graphs.Existing methods primarilyfocus onstaticknowledge graphs，failing tofulluizetheabundanttemporal informationpresentinlarge-scaleknowledgestructures，andneglectingpotentialneighbors thatarenotdirectly conectedwhileintegratingentityneighborhood information.Toaddressthesechallenges，thispaperproposedanentityalignmentmodel，EA，totegateighborhoodiformationintomporalnowledgeaphs.irstly，itonstructede ralfeaturesofentitiesbycapturingtheiractivitiesatdiferentimepoints.Next，itintroducedaneighborhoodsimilaritymeasurement methodto identifypotential neighboring nodes，enabling message pasing acrossseparatechannelstoextractdiverse neighborhoodrepresentationsandcapturedomain-specificfeatures.Finallyitcombined the temporalandneighborhoodfeatures of ntitiestogenerateenrichedentityembeddings.Experimentalresultsdemonstratethatthe proposed modelimproves hits@1 by1.8，1.7，1.7，and1.9 percentagepointson fourreal-worlddatasets，DICEWS-1K/20andYAGO-WIKI50K5K/1K，compared tothebest-performing baselinemodels.Thesefindings validatetheefectivenessof the proposedapproach. Key words：temporal knowledge graph;entity alignment;knowledge embeddng;entity activity；potential neighbors

0 引言

知识图谱（knowledgegraph，KG）[1]以图结构的形式存储知识，图中的节点表示实体，节点之间的边表示关系。（剩余19392字）