Knowledge graph recommendation method based on preference migration

The invention belongs to the technical field of recommendation systems, and provides a knowledge graph recommendation method based on preference migration. The attention embedding propagation is mainly composed of an attention embedding propagation layer, a preference migration layer and a prediction layer. Wherein the attention embedding propagation layer recursively propagates the embedding of neighbors of nodes on the collaborative knowledge graph to refine the embedding of the nodes, learns the weight of each neighbor in the propagation process through a knowledge perception attention mechanism, and finally aggregates user and project representations from all layers; the preference migration layer finds a user belonging to the cold start through a set threshold value, then carries out secondary propagation on neighbor users of the cold start user on a user logic interaction space based on a graph convolutional network architecture, and uses a normalized intersection-union ratio to represent the weight of