Balancing Augmentation With Edge Utility Filter for Signed Graph Neural Networks

Many real-world networks are signed networks containing positive and negative edges. The existence of negative edges in the signed graph neural network has two consequences. One is the semantic imbalance, as the negative edges are hard to obtain though they may potentially include more useful information. The other is the structural unbalance, e.g., unbalanced triangles, an indication of incompatible relationship among nodes. This paper proposes a balancing augmentation to address the two challenges. Firstly, the utility of each negative edge is determined by calculating its occurrence in balanced structures. Secondly, the original signed graph is selectively augmented with the use of (1) an edge perturbation regulator to balance the number of positive and negative edges and to determine the ratio of perturbed edges and (2) an edge utility filter to remove the negative edges with low utility. Finally, a signed graph neural network is trained on the augmented graph. The theoretical analysis is conducted to prove the effectiveness of each module and the experiments demonstrate that the proposed method can significantly improve the performance of three backbone models in link sign prediction task, with up to 22.8% in the AUC and 19.7% in F1 scores, across five real-world datasets.