Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification

Benchmark Model Rank Results
node-classification-on-amazon-computers-1GAT#1Accuracy: 94.09±0.37
node-classification-on-amazon-computers-1GCN#2Accuracy: 93.99±0.12
node-classification-on-amazon-computers-1GraphSAGE#3Accuracy: 93.25±0.14
node-classification-on-amazon-photo-1GraphSAGE#1Accuracy: 96.78 ± 0.23
node-classification-on-amazon-photo-1GAT#2Accuracy: 96.60 ± 0.33
node-classification-on-amazon-photo-1GCN#3Accuracy: 96.10 ± 0.46
node-classification-on-amazon-ratingsGAT#1Accuracy (%): 55.54 ± 0.51
node-classification-on-amazon-ratingsGraphSAGE#2Accuracy (%): 55.40 ± 0.21
node-classification-on-amazon-ratingsGCN#4Accuracy (%): 53.80 ± 0.60
node-classification-on-citeseer-with-publicGCN#20Accuracy: 73.14± 0.67
node-classification-on-coauthor-csGraphSAGE#3Accuracy: 96.38±0.11
node-classification-on-coauthor-physicsGCN#3Accuracy: 97.46 ± 0.10
node-classification-on-cora-with-public-splitGCN#6Accuracy: 85.1 ± 0.7
node-classification-on-minesweeperGCN#1AUCROC: 97.86 ± 0.24
node-classification-on-minesweeperGraphSAGE#2AUCROC: 97.77 ± 0.62
node-classification-on-minesweeperGAT#3AUCROC: 97.73 ± 0.73
node-classification-on-pokecGCN#2Accuracy: 86.33 ± 0.17
node-classification-on-pubmed-with-publicGCN#8Accuracy: 81.12 ± 0.52
node-classification-on-questionsGCN#2AUCROC: 79.02 ± 0.60
node-classification-on-roman-empireGCN#4Accuracy (% ): 91.27±0.20
node-property-prediction-on-ogbn-arxivGCN#34Test Accuracy: 0.7360 ± 0.0018Ext. data: No
node-property-prediction-on-ogbn-arxivGraphSAGE#44Test Accuracy: 0.7295 ± 0.0031Ext. data: No
node-property-prediction-on-ogbn-productsGraphSAGE#23Test Accuracy: 0.8389 ± 0.0036Ext. data: No
node-property-prediction-on-ogbn-productsGCN#30Test Accuracy: 0.8233 ± 0.0019Ext. data: No
node-property-prediction-on-ogbn-proteinsGAT#14Ext. data: NoTest ROC-AUC: 0.8501 ± 0.0046
node-property-prediction-on-ogbn-proteinsGraphSAGE#16Ext. data: NoTest ROC-AUC: 0.8221 ± 0.0032