无线电通信技术2025，Vol.51Issue(3)：493-500,8.DOI:10.3969/j.issn.1003-3114.2025.03.008
基于关系感知语言图的知识查询网络
Knowledge Query Network Based on Relation-Aware Language Graph
摘要
Abstract
Question Answering(QA)is a task that requires reasoning about natural language context,existing work almost uses Graph Neural Network(GNN)to enhance Language Model(LM)to encode Knowledge Graph(KG)information.However,most GNN based QA modules do not utilize rich relational information of KG and rely on limited information interaction between LM and KG.To solve these prob-lems,a knowledge query network based on Relation-Aware Language Graph(RALG)is proposed,which conducts joint reasoning of entity re-lation language and graph in a unified way.RALG builds a meta path tag that learns embeddings based on different structural and semantic relationships.Then,a Relational Aware Self-Attention(RASA)module integrates different modes through cross-modal relative position devia-tion,and guides the information exchange between different modal related entities.Performance advantages of RALG are evaluated on general knowledge QA datasets(CommonsenseQA and OpenBookQA)and medical QA datasets(MedQA-USMLE).关键词
图神经网络/语言模型/知识图谱/关系感知/跨模态Key words
GNN/LM/KG/relation-aware/cross-modal分类
信息技术与安全科学引用本文复制引用
陶薇薇,王延红..基于关系感知语言图的知识查询网络[J].无线电通信技术,2025,51(3):493-500,8.基金项目
四川省科技计划资助(2019JDPT0009) Sichuan Science and Technology Program(2019JDPT0009) （2019JDPT0009）
