Recipe conception method based on relational memory-cascade transfer

The invention relates to the technical field of neural networks, and relates to a recipe conception method based on relation memory-cascade transfer, and the method comprises the steps: building a recipe data set based on a Chinese menu, building a relation memory-cascade transfer model in the aspect of food material selection, improving the accuracy of a model for predicting the affinity of food materials, and enabling the matching of the food materials to be more reasonable; according to the method and the system, the recipe is generated according to the previously selected food materials by using ChatGPT, so that the prompt method for the existing recipe is effectively designed, and the tedious process caused by using other generative models is greatly reduced. Different from previous research contents, the method is driven by local data of China and focuses on matching of local food materials, and a more delicious new Chinese menu is created. 本申请涉及神经网络技术领域，关于一种基于关系记忆-cascsded transformer的食谱构思方法，步骤包括：基于中国菜