Research Article
Towards Revealing Parallel Adversarial Attack on Politician Socialnet of Graph Structure
Algorithm 1
Parallel fast gradient attack (P-FGA).
| | Input: , attack target set , perturbation budget | | | Output: perturbation set | | (1) | Train the GCN model on original graph | | (2) | Initialize | | (3) | Initialize perturbation set | | (4) | for to do | | (5) | //GCN-based Gradient Computation | | (6) | Calculate multi-node target loss function | | (7) | Construct based on the : | | | , , | | | | | (8) | //Perturbation Selection | | (9) | Select where , having the maximum | | (10) | //Perturbation Execution | | (11) | Obtain the adjacency matrix by | | (12) | Generate a new adversarial graph | | (13) | Add to | | (14) | end | | (15) | return |
|