计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (22): 329-342.DOI: 10.3778/j.issn.1002-8331.2303-0400

• 工程与应用 • 上一篇    

面向食品贮藏领域的知识图谱构建方法研究

辛辉,谢镇玺,李朋骏,王金龙,熊晓芸   

  1. 青岛理工大学 信息与控制工程学院,山东 青岛 266525
  • 出版日期:2023-11-15 发布日期:2023-11-15

Research on Knowledge Graph Construction Method for Food Storage Field

XIN Hui, XIE Zhenxi, LI Pengjun, WANG Jinlong, XIONG Xiaoyun   

  1. School of Information and Control Engineering, Qingdao University of Technology, Qingdao, Shandong 266525, China
  • Online:2023-11-15 Published:2023-11-15

摘要: 食品贮藏是人们追求高质量饮食的关键环节,也是建设资源节约型社会的着力点,而海量多源异构的食品数据存在信息过载、缺乏系统化表示和查询困难等问题,制约了相关从业人员的研究开发和普通居民获取信息的有效性。基于知识图谱结构化表达语义信息的特点,将其与食品贮藏相融合,设计出一套完整的食品贮藏领域知识图谱构建框架,为食品研究和应用提供新视角和智能化方法。首先针对多源异构的食品贮藏数据进行分析,提出了多元概念间关系的表达模式,用以弥补图谱三元组知识表示的缺陷,进而完成模式层构建;然后重点对非结构化知识采取改进的融合多特征的实体识别方法完成实体抽取,并用多分类模型和提出的多元关系抽取算法进行关系抽取;针对抽取产生的知识冗余,利用基于词典和相似度匹配方法完成融合;最后以Neo4j图数据库完成存储。整个方法框架的设计通过实验进行了合理性和有效性验证,并对构建结果进行可视化分析与语义检索。

关键词: 食品贮藏, 知识图谱, 特征融合, 多元关系表示, 知识融合

Abstract: Food storage is the key link in people’s pursuit of high-quality diet and the focus of building a resource-saving society. However, there are problems of information overload, lack of systematic representation and difficulty with querying in the huge amount of multi-source heterogeneous food data, which restrict the research and development of relevant practitioners and the effectiveness of information access for the general population. Based on the characteristics of structurized semantic information expressed in the knowledge graph, this paper integrates it with food storage and designs a complete framework for the construction of knowledge graph in the field of food storage, which provides new perspectives and intelligent methods for food research and application. Firstly, according to the characteristics of multi-source and heterogeneous food storage data and ontology construction method, this paper proposes the expression model of multi-entity relationship, which makes up for the deficiency of knowledge representation of triples, and completes the schema layer construction. Secondly, this paper focuses on the unstructured knowledge by using an improved named entity recognition method that combines multiple features to complete the entity extraction, and uses the multi-classification model and the proposed multi-relationship extraction algorithm to complete the relationship extraction. Knowledge fusion is completed by using the methods of dictionary-based and similarity matching. Finally, the Neo4j is applied to the database stores. The rationality and effectiveness of the entire framework are verified by experiments. Moreover, visual analysis and semantic search are completed on the successfully constructed knowledge graph.

Key words: food storage, knowledge graph, feature fusion, multi-relation representation, knowledge fusion