面向产品指标图谱的知识表示学习方法研究  

Knowledge Representation Learning Method for Product Index Graph

作　　者：邱雪 单世民[1] 魏宏夔 王恺[1] 杨念顺 QIU Xue;SHAN Shimin;WEI Hongkui;WANG Kai;YANG Nianshun(Liaoning Key Laboratory of Ubiquitous Network and Service Software,Dalian University of Technology,DaLian 116620,China;State Key Laboratory of Intelligent Manufacturing System Technology,Beijing Institute of Electronic System Engineering,Beijing 100854,China)

机构地区：[1]大连理工大学辽宁省泛在网络与服务软件重点实验室,辽宁大连116620 [2]北京电子工程总体研究所复杂产品智能制造系统技术国家重点实验室,北京100854

出　　处：《山西大学学报（自然科学版）》2022年第4期894-901,共8页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(61672128);中央高校基本科研业务费专项资金(DUT20TD107);复杂产品智能制造系统技术国家重点实验室开放基金课题。

摘　　要：通过知识表示学习方法将产品和指标表示为低维分布式向量,为后续的产品指标补全和产品设计方案预测奠定基础。然而现有的知识表示学习方法只处理实体-关系之间的离散型关联,而对于数值型指标的研究尚属空白。为此,文章依托复杂产品指标图谱的实际业务需求,设计了产品数值型指标的表示学习策略。针对关系三元组和指标三元组的语义差异,探索全新的联合学习训练方案。文章在五个前沿知识表示学习算法上进行实验,其中基于依次学习训练方案的ConvE算法,在图谱链接预测任务上Hit@10指标达到了最优的90.27%。实验结果验证了本文数值型指标表示方法和联合训练方案的有效性。Through the knowledge representation learning method,the products and indicators are expressed as low-dimensional distributed vectors,which lays on the foundation for the subsequent product index completion and product design scheme prediction.However,the existing knowledge representation learning methods only deal with the discrete association between entities and relationships,while the research on numerical index is still blank.Therefore,based on the actual business requirements of complex product index graph,this paper designs the representation learning strategy of product numerical index.Aiming at the semantic differences between relational triples and index triples,a new joint learning training scheme is explored.The experiments are carried out on five cutting-edge knowledge representation learning algorithms,among which ConvE algorithm based on sequential learning scheme whose Hit@10 reached the optimal 90.27%in knowledge graph link prediction task.The experimental results verify the effectiveness of the numerical index representation method and joint training scheme.

关 键 词：知识表示学习 知识图谱 数值实体表示 产品指标预测 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]