Home

Awesome

Robust Graph Clustering via Meta Learning for Noisy Graphs

MetaGC is a GNN-based graph clustering method that is robust against noise edges. MetaGC consists of a GNN-based clustering model, and a meta model that adaptively adjusts the weights of node pairs. MetaGC uses a modularity-based loss function with theoretical justification. In our experiments, we demonstrate that MetaGC learns weights effectively and thus outperforms the state-of-the-art GNN-based competitors, even when they are equipped with separate denoising schemes, on five real-world graphs under varying levels of noise.

Environments

important argments

Running MetaGC with Cora graph with noise level I

python3 main.py --graph_name cora --noise_level 1 --batch_size 128 --num_hiddens 64 --num_epochs 200 --max_epochs 1500 --num_patience 50 --c_lr 0.001 --m_lr 0.005