Abstract: [Purpose/Significance] This study intends to study the single point early warning mechanism based on the fusion of multi-source data and the diffusion early warning mechanism based on correlation analysis to improve the food safety risk early warning and efficiently promote innovation in food safety supervision. [Design/Methodology] This paper firstly uses multi-layer perceptron and word vector model to learn and fuse multi-source data to achieve single point early warning decision-making for the currently sampled business subjects. In addition, correlation analysis is adopted to diffuse the warning information according to the structured relationship between business subjects, in order to achieve real-time early warning decision-making for business subjects related to the currently sampled business subjects. [Findings/Conclusion] By using the decision-making mechanisms of single point early warning and diffusion early warning, the risk warning level can be accurately located, and the early warning can be diffused, thus creat the basic conditions for the establishment of a real-time early warning regulatory mechanism and a closed-loop management mechanism for early warning disposal. [Originality/Value] The single point early warning mechanism can efficiently integrate information from many different sources to realize precise single point early warning; the diffusion early warning mechanism can automate the diffusion from single point to multiple points early warning based on correlation analysis, realizing efficient early warning from point to surface. On the whole, the early warning mechanism proposed in this paper can improve the efficiency and quality of food safety supervision.

Keywords: Food safety risk warning, Multi-source data fusion, Correlation analysis, Single point early warning, Diffusion early warning