化妆品风险物质知识图谱构建及应用  

Construction and application of knowledge graph forcosmetic risk substances

作　　者：赵敏 毛典辉[1,2] 张青川 吕东东 刘一铭 陈俊华 ZHAO Min;MAO Dian-hui;ZHANG Qing-chuan;LYU Dong-dong;LIU Yi-ming;CHEN Jun-hua(Institute of Cosmetic Regulatory Science,Beijing Technology and Business University,Beijing 100048,China;Beijing Key Laboratory of Big Data Technology for Food Safety,School of Computer Science and Engineering,Beijing Technology and Business University,Beijing 100048,China;Institute of Standardization Theory and Strategy,China National Institute of Standardization,Beijing 100088,China)

机构地区：[1]北京工商大学国家药品监督管理局化妆品监管科学研究基地,北京100048 [2]北京工商大学计算机学院食品安全大数据技术北京重点实验室,北京100048 [3]中国标准化研究院标准化理论与战略研究所,北京100088

出　　处：《计算机工程与设计》2023年第12期3784-3793,共10页Computer Engineering and Design

基　　金：北京市社会科学基金项目(19GLB036);国家社会科学基金项目(18BGL202)。

摘　　要：针对国内外化妆品风险物质多语言特性和复杂关联的特点,提出一种基于双通道图神经网络的邻域匹配算法。采用图神经网络学习实体属性特征和跨域交互特性,将不同特性实体映射到相同的向量空间,通过邻域匹配网络聚合实体邻域特征,为每个实体构建邻域网络以实现实体对齐,并应用于多语言风险物质知识图谱及问答系统构建。实验结果表明,在化妆品风险物质数据集上该方法获得的Hits@1、Hits@10与MRR值都优于其它基线模型,分别平均提升6.37%、8.17%与9.37%。Aiming at the multi-lingual characteristics and complex association of domestic and foreign cosmetic risk substances,a neighborhood matching algorithm based on two-channel graph neural network coding was proposed.To map entities with diffe-rent properties to a unified vector space,the graph neural network was proposed to learn entity attributes and cross-domain inte-raction characteristics.The neighborhood matching network was employed to aggregate the neighborhood characteristics of entities,and the neighborhood network was constructed for each entity to achieve entity alignment.It was applied to construct multi-lingual risky substance knowledge graph and question and answer system.Experimental results show that the Hits@1,Hits@10 and MRR values obtained using this method on the cosmetic risk substances dataset are better than those obtained using other baseline models,with an average increase of 6.37%,8.17%and 9.37%,respectively.

关 键 词：化妆品风险物质 知识图谱 跨语言实体对齐 贝叶斯算法 图卷积神经网络 邻域匹配网络 问答系统 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]