Research on multi-hop reasoning question and answer model for foodborne disease incidents

This paper addresses the problem that there is little research on question-and-answer technical reasoning in the field of food safety and the difficulty of mining deep information for related reasoning. This paper proposes a multi-hop reasoning model for food safety incident knowledge graphs. Firstly, a knowledge graph triad network is established by extracting and embedding the knowledge related to food accidents. Secondly, a two-module system is established by combining the two-channel theory. Module 1 adopts the attention mechanism to calculate the weight coefficients on the relational edges and expand them, and module 2 adopts the GAT neural network to infer and calculate the hidden representations of entities, and the inference prediction of the tail entities is achieved through the interactive iteration of module 1 and module 2. Experiments show that after iterative training of the model on food safety data, the correct answer can be inferred with good accuracy. The multi-hop based food safety incident inference model is highly accurate and interpretable, and can be applied to the question and answer system to assist relevant personnel to have a quick query and determination on the causes, influencing factors, causative factors and their characteristics, pathogenic mechanisms and clinical manifestations of food safety incidents to reduce the occurrence of incidents.