The invention provides a graph data representation method based on multiple virtual nodes and random hash propagation, and relates to the field of graph data representation, and the method comprises the steps: carrying out node importance evaluation and feature similarity clustering on an original graph, and dividing the nodes into a plurality of node clusters; introducing at least one virtual node for each node cluster, constructing a connection relationship between the virtual node and all real nodes in the cluster, and forming a structure enhancement graph; maintaining a sketch vector of which the fixed dimension is K for each real node and each virtual node on the structure enhancement graph; based on a random
hash function family, performing multi-round propagation and updating on the sketch vector, and generating node representation fusing node attributes and multi-
order structure information; and directly applying the node representation to a graph
data processing task. According to the technical scheme, joint modeling of the node attribute information and the multi-
order structure information in the complex graph structure is achieved.