Research On Construction And Completion Technology Of Nutrition Knowledge Graph

In recent years,social economy develops rapidly and people's health awareness improves.People pay more attentions to the diet balance and physical health.Diet is one of the most important things in people's daily life.Based on scientific nutrition knowledge,providing food and nutrition suggestions for people is important.The rapid development of big data and artificial intelligence technology provides a new perspective for nutrition research.This subject constructs the nutrition knowledge graph based on nutrition data in professional nutrition books and websites,and studies knowledge graph completion methods.We have carried out the following work:(1)Firstly,this paper constructs the concept and relation system of nutrition knowledge.This paper gives the definitions of entity types and relation types in the knowledge graph and gives relevant examples.The knowledge graph contains eight entity types and twelve relation types.In this paper,we obtain nutrition data from professional nutrition books and websites.By performing named entity recognition,entity alignment,and knowledge representation on these data,this paper constructs the nutrition knowledge graph based on the word data.(2)Then,this paper studies the relation construction of nutrition knowledge graph.The data of knowledge graph is always incomplete.In order to complete the missing relation between entities,this paper presents the relation classification model of nutrition entity.We use the R-GCN model and multi-attention mechanism to extract the nutrition entity features and to classify and predict the potential relation between entities.The experiment compares with the relation classification models based on traditional machine learning methods,and the results show the advantages of the proposed method.(3)Finally,this paper proposes a method to complete the nutrition knowledge graph.We study link prediction methods based on knowledge representation for fully considering the relationship in triples and for filling incomplete triples.We make statistics and analysis of the nutrition data characteristics,and use the Trans E and Rotat E models to represent the nodes in the nutrition knowledge graph,and carry out link prediction experiments.In addition,this paper also discusses the influence of different negative sampling techniques on link prediction results.In summary,this paper focuses on the field of nutrition,and deeply studies the construction and completion of nutrition knowledge graph.It paves the way for diet recommendation research based on the nutrition knowledge graph.