Social robot detection model and method based on graph contrast learning

The invention belongs to the technical field of social robot detection, and particularly relates to a social robot detection model and method based on graph comparative learning, and the model comprises an information coding module, a data enhancement module, a comparative learning module and a node classification module. The information coding module constructs a social relation topological graph, and carries out vectorization operation on semantic features and attribute features of accounts to obtain initial representation vectors of nodes; the data enhancement module is used for augmenting the constructed social relation topological graph through a plurality of data enhancement modes to generate a plurality of views conforming to original data distribution; the contrast learning module encodes the plurality of augmented views by using a graph neural network, and obtains node representation with the maximum convergence through minimizing contrast loss; and the node classification module predicts a node labe