Studies On The Food Safety Knowledge Graph Based On Graph Neural Networks

With the development of economy,people’s requirement for the quality of life has gradually changed from the need for survival to higher standards and spiritual needs.Food safety is an important issue for people’s livelihood.Once it appears,it will attract wide attention.As the food production industry chain is long and varied,the whole chain supervision of food production and circulation is facing great challenges.At the same time,different food standards make it difficult to query and retrieve specific food standard parameters.In the wake of developments in science and technology,knowledge graph provides a new idea for constructing a perfect food safety standard retrieval system.This knowledge structure integrates complex and disordered information into more logical knowledge representation.Based on relevant national standards in the field of food safety,this paper constructed a food safety knowledge graph and proposed the relevant knowledge graph completion method.The main work is as follows:1.By extracting the contents of national food safety standard documents from three data sources,which are food categories,food additives and pesticide residues in food limited,through a series of data processing operations including data cleaning and formatting,we can get valid triples which can be applied in the knowledge graph.At the same time,the scheme layer of the food safety knowledge graph is built,a knowledge graph oriented to the contents of food safety standards is constructed,and a visual query system is designed,which can be easy for users to query and display.2.In order to solve the problem of missing entity types in the incomplete knowledge graph,this paper proposes a graph neural network architecture that fuses graph structure information and entity description information,and analyzes the influence of different feature fusion methods on entity classification through experiments.Experiment results show that the proposed method can improve the accuracy of classification prediction greatly compared with the single relational graph convolution network model.3.In order to solve the problem of missing triples of incomplete knowledge graph,this paper proposes a graph neural network link prediction model based on encoder-decoder,and analyzes the influence of different decoder designs on link prediction results combined with comparative experiments.Experiments show that compared with the single relational graph convolutional network model,the link prediction model architecture in this paper has better performance and has significantly improved in several evaluation indicators.The food safety knowledge graph system can provide a better knowledge base and retrieval method for many food safety fields,which include food quality management,food traceability and detection.And this system can also provide a new solution for the informational and intelligent food safety management.