Imbalanced node classification with Graph Neural Networks: A unified approach leveraging homophily and label information.pptx
1. Van Thuy Hoang
Network Science Lab
Dept. of Artificial Intelligence
The Catholic University of Korea
E-mail: hoangvanthuy90@gmail.com
2023-11-13
Dingyang Lv et. al.; A.S.C 23
2. 2
Graph Convolutional Networks (GCNs)
Generate node embeddings based on local network neighborhoods
Nodes have embeddings at each layer, repeating combine messages
from their neighbor using neural networks
3. 3
phenomenon
Many real-world graphs: as neighbor class imbalance, which is characterized by
frequent connections between dissimilar nodes, a scenario reflecting low
homophily
Classical GNNs tend to overlook this issue, leading to a significant decline in
performance.
4. 4
phenomenon
the key issues of Neighbor Class Imbalance: Supposing target node 𝑣 is a fraud
object, its neighbors in the dashed circle are class imbalanced since benign
nodes are significantly more than fraud ones
Class-imbalanced Learning
13. 13
Conclusion
a novel message passing based on homophily degree estimation and label
propagation to enhance the quality of representation learning for the classic
GNNs.
three key techniques:
a homophily degree estimation
label propagation enhanced representation
high-order neighborhood information aggregation