面向食品监管领域的知识图谱构建研究  

Research on Knowledge Graph Construction for Food Supervision

作　　者：卜意磊 庞文迪 吴甜甜 杜奕坤 李珊 BU Yi-lei;PANG Wen-di;WU Tian-tian;DU Yi-kun;LI Shan(Jiangsu Provincial Administration for Industry and Commerce Information Center,Nanjing 210019,China;School of Information,Beijing University of Chemical Technology,Beijing 100029,China;School of Economics and Management,Nanjing University of Aeronautics and Astronautics,Nanjing 211100,China)

机构地区：[1]江苏省工商行政管理局信息中心,江苏南京210019 [2]北京化工大学信息学院,北京100029 [3]南京航空航天大学经济与管理学院,江苏南京211100

出　　处：《计算机技术与发展》2023年第6期202-207,214,共7页Computer Technology and Development

基　　金：国家自然科学基金面上项目(72174086)。

摘　　要：该文提出了面向食品监管领域的知识图谱构建研究,通过整理食品监管领域的相关文件和政策,并进行实体识别、实体关系识别、实体对齐构建食品监管领域知识图谱。其中基于双向长短时记忆网络与条件随机场结合的BiLSTM-CRF模型进行实体识别,准确率达0.96;基于食品监管实体的归类结果,确定同标签的实体间的分类关系,并创建“文本-实体”矩阵,提取出包含某实体对的所有句子,归纳实体对之间的关系;通过聚类进行实体对齐,并基于Neo4j存储和呈现图谱。构建的食品监管知识图谱弥补了食品监管领域知识图谱研究的空白,提升了食品监管体系和监管能力现代化水平。Research on knowledge graph construction for the field of food supervision is proposed.By sorting out relevant documents and policies in the field of food supervision,and performing entity identification,entity relationship identification and entity alignment,a knowledge graph in the field of food supervision is constructed.Among them,the BiLSTM-CRF model based on the combination of bidirectional long short-term memory network and conditional random field is used for entity recognition,with an accuracy rate of 0.96.Based on the classification results of food regulatory entities,the classification relationship between entities with the same label is determined,and a"text-entity"matrix is constructed to extract all sentences containing a certain entity pair and summarize the relationship between entity pairs.Entity alignment is performed through clustering,and graphs based on Neo4j are stored and presented.The food supervision knowledge map constructed fills the gap of knowledge map research in the field of food supervision,and improves the modernization level of the food supervision system and supervision capacity.

关 键 词：知识图谱 食品监管 实体识别 关系抽取 实体对齐 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]