.png){kind=icon}
.png){kind=icon}
Return
Augmented Graph Convolutional Network for Enhancing Label Reachability
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2025.3555997.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph Convolutional Networks (GCNs) have emerged as a leading approach for semi-supervised node classification. However, due to the uneven distribution of labeled nodes in graphs, only a limited subset of unlabeled nodes can directly access the labeled nodes. This disconnection between the labeled and unlabeled nodes significantly hinders GCNs performance. To address this challenge, we propose an Augmented Graph Convolutional Network for enhancing Labeled Reachability (AGCN-LR), which is designed to improve information flow to low-degree nodes by strengthening their connectivity with labeled nodes. The core of AGCN-LR comprises two parts. First, we selectively connect low-degree nodes to labeled nodes by calculating feature similarity and shortest path distances to generate augmented graphs, thereby enhancing label reachability. Second, we leverage GCNs to learn these augmented graphs and integrate embeddings from these graphs through an attention mechanism to bolster overall performance. Furthermore, to capture consistent information across augmented graphs, we incorporate a tailored contrastive loss function, facilitating consistent contextual learning across different augmented graphs. Experimental results demonstrate that AGCN-LR substantially enhances performance in semi-supervised node classification tasks across various benchmark datasets, exhibiting stronger advantages in sparse label and noisy graph scenarios.
Keywords:
Graph convolutional networks
Symmetric matrices
Robustness
Layout
Contrastive learning
Adaptation models
Semantics
Noise measurement
Noise
Generators
label reachability
contrastive learning
node classification
semi-supervised learning
Journal
IF:
3.6
Papers:
9.7W
Citations:
29.4W