.png){kind=icon}
.png){kind=icon}
Return
GraphBSSN: A simple yet effective generative method for node classification in class-imbalanced graphs
.png){kind=icon}
PRE
AI
DOI：10.1016/j.knosys.2025.114175.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph Neural Networks (GNNs) have shown significant promise in node classification tasks. However, in practical scenarios, the distribution of samples across various categories is often imbalanced. When GNNs are trained on such imbalanced graphs, they often fail to adequately represent the minority classes, resulting in a severe decline in performance. To address this issue, one potential solution is to generate supplementary samples for the minority classes to balance the graph. However, the existing generative methods are complex and tend to squeeze the subspace of minority classes. Additionally, the topology of synthesized samples deviates from the true neighbor distribution. In response, we propose a simple yet effective generative framework, GraphBSSN, by synthesizing Boundary Samples and sampling Similar Nodes. Specifically, we devise a boundary-aware feature synthesis strategy to expand the decision boundary of minority classes. Moreover, we design a similar node-based topology modeling method to position the synthesized samples within a reasonable distribution of neighbors. Our experimental results on eight class-imbalanced datasets demonstrate the effectiveness of the proposed method. The datasets and codes are available on GitHub at https://github.com/ZZY-GraphMiningLab/GraphBSSN .
Keywords:
Graph Neural Networks
Node Classification
Class Imbalance
Sample Generation
Boundary-aware Synthesis
Similar Node Sampling
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W
Organization
No organization information available