Abnormal User Detection via Multiview Graph Clustering in the Mobile e-Commerce Network

Algorithm 1

Implementation of the DM-VCDM.

Input: user-similarity graph , number of iterations , number of clusters ;

Output: The abnormal user detection results;

Initialize the parameter of the GCNs , and = =1/2;

for =1 to 2:

  Encode to get latent embedding by Equation (3)

  Initialize cluster assignments and using K-means++ in each view;

  Initialize and by Equation (5) and Equation (6) respectively;

end for;

for =1 to :

  for =1 to 2:

   Update and using backward propagate SGD according to Equation (10) and

  Equation (11);

   Update latent embedding by Equation (3);

   Update and by Equation (5) and Equation (6) respectively;

  end for;

  Calculate according to Equation (7);

  Update and by Equation (12);

end for;

Calculate the detection results by Equation (7);

Return cluster assignments for all users