A hierarchical network embedding method based on network partitioning

The invention discloses a hierarchical network embedding method based on network partitions, which comprises the following steps: judging an affiliated community of nodes in an original network graphbased on modularity gain, and determining the number of network partitions; setting the number of the network partitions as a scale threshold of the minimum-scale abstract graph, performing graph abstraction on the original network graph based on a hybrid collapsing method, and outputting the abstract graph of which the scale is gradually reduced until the thickest abstract graph is equal to the scale of the minimum-scale abstract graph; performing representation learning on the thickest abstract graph according to a baseline algorithm to obtain representation of the thickest abstract graph; and propagating and refining the representation of the thickest abstract graph layer by layer through an embedded propagation method to obtain the representation of the original graph. According to thehierarchical network embedding method, the network representation effect of the baseline method is improved, and the human intervention problem of threshold setting is reduced, and the accuracy of network representation is further improved, and the hierarchical network embedding method can be widely applied to the fields of link prediction, multi-label node classification, community discovery andrecommendation systems.