融合实体特征聚合和关系语义聚合的推理模型

打开文本图片集

关键词：时序知识图谱；图结构；实体特征聚合；关系语义聚合

中图分类号：TP391.1 文献标志码：A 文章编号：1001-3695（2025）10-014-3005-07

doi：10.19734/j.issn.1001-3695.2024.11.0537

Reasoning model integrating entity feature aggregation and relation semantic aggregation

Dong Wenyong1，2a，Liang Zhixue2a， Zhou Mengqiang2a， Jia Yajie²b （1.SchooloffatiekSuityingUnesitficlce&LauskeXi84na;lf ComputerScience，b.SchoolofCyber Science&Engineering，Wuhan University，Wuhan43oo72，China）

Abstract：Mostexisting temporal knowledgegraph reasoning modelsrelyonrelational graph neural networks tocapturesemanticdependencies between entities ineachsnapshot.To beterutiizestructural information within graph data，this paper proposed the EFRSA reasoning model，which integratedentityfeatureaggregationandrelational semanticaggegation.This model effetivelycaptured semanticdependencies amongconcurent entities at eachtimestamp.Through its entityfeatureaggregationmodule，EFRSA identifiedandleveragedthepotential significantassciations amongco-occurrng entities.Aditionaly，EFRSAintroducedarelationsemanticagregatiomodulebasedonelatioalsubgaphassociationstofullexpresrelational semantic information in the graph structure.Experimental resultsondatasets suchas ICEWS14，GDELT，YAGO，and WIKI show that EFRSA achieves an MRR improvement of 0.89～3.24 in entity prediction and outperforms other methods in relation semantic prediction，thereby enhancing the model's reasoning capability.

Key words：temporal knowledge graph； graph structure;entity feature agregation；relationsemanticaggregation

0 引言

近年来，随着信息技术的快速发展和大数据的广泛应用，互联网上的信息量呈爆炸式增长，这使得人们面临如何有效组织、管理和利用这些庞大信息的挑战。（剩余17154字）