Attribute network abnormal node detection method based on personalized PageRank and comparative learning

The invention discloses an attribute network abnormal node detection method based on personalized PageRank and comparative learning, and belongs to the field of abnormal node detection. An adaptive sampling strategy based on PPR is provided, an adaptive subgraph scale is selected according to centrality of different nodes in an attribute network, and local structure information is obtained, so that the problem that the detection accuracy is reduced due to context information loss or noise introduction caused by a fixed sampling strategy is solved. A K-nearest neighbor algorithm is utilized to independently search a node nearest neighbor of an attribute network from an attribute space and obtain global attribute information, and abnormal information additionally provided by attribute features of nodes is fully captured. Two contrast normal forms are constructed from two aspects of a local structure and a global attribute, a contrast learning loss function and an abnormal value calculation formula directly aimi