Clarify Confused Nodes via Separated Learning

In this paper, we first propose a new metric, termed Neighborhood Confusion (NC), to facilitate a more reliable separation of nodes. These pave the way for Neighborhood Confusion-guided Graph Convolutional Network (NCGCN), in which nodes are grouped by their NC values and accept intra-group weight sharing and message passing.

1 Overview of Model Architecture and Performance

2 Python environment setup with Conda

conda env create -f requirement.yaml

3 Data Preparation

For all datasets, we adopt dense data splitting as in [20], i.e., randomly split them into training, validation and testing sets with a proportion of 60%/20%/20%.

4 Code Execution

4.1 File Structure

• data -- put data in this dir
• model -- GCN(NCGCN) model
• index -- index of nodes' neighborhood
• main.py -- run this
• utils.py -- utils

4.2 Code Execution

python main.py --dataset computers --model NCGCN --device 3 --lr 5e-2 --wd 5e-5 --dp1 0.2 --dp2 0.6 --hops 2 --addself 0 --threshold 0.7 --hidden 512

4.3 Hyper-parameter Space

• learning rate: [1e-3, 5e-3, 1e-2, 5e-2, 0.1]
• weight decay: [0, 5e-5, 1e-4, 5e-4, 1e-3]
• dropout rate: [0, 0.9] with 0.1 interval
• hop k: [1, 2]
• Threshold T: [3, 4, 5, 6, 7]
• Add self-loop or not: [True, False]

5 Citation

If you find this work useful, please cite our paper:

@article{zhou2025ncgnn,
  title={Clarify Confused Nodes via Separated Learning},
  author={Jiajun Zhou, Shengbo Gong, Xuanze Chen, Chenxuan Xie, Shanqing Yu, Qi Xuan and Xiaoniu Yang},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2025}
}