Two Sides of the Same Coin: Heterophily and Oversmoothing in Graph Convolutional Neural Networks

Benchmark Model Rank Results
node-classification-on-actorGGCN#15Accuracy: 37.54 ± 1.56
node-classification-on-chameleonGGCN#16Accuracy: 71.14 ± 1.84
node-classification-on-citeseer-48-32-20GGCN#91:1 Accuracy: 77.14 ± 1.45
node-classification-on-cora-48-32-20-fixedGGCN#141:1 Accuracy: 87.95 ± 1.05
node-classification-on-cornellGGCN#13Accuracy: 85.68 ± 6.63
node-classification-on-non-homophilic-10GGCN#61:1 Accuracy: 37.54 ± 1.56
node-classification-on-non-homophilic-10GPRGCN#211:1 Accuracy: 35.16 ± 0.9
node-classification-on-non-homophilic-11GGCN#91:1 Accuracy: 71.14 ±1.84
node-classification-on-non-homophilic-12GGCN#161:1 Accuracy: 55.17 ± 1.58
node-classification-on-non-homophilic-7GGCN#61:1 Accuracy: 85.68 ± 6.63
node-classification-on-non-homophilic-8GGCN#151:1 Accuracy: 86.86 ± 3.29
node-classification-on-non-homophilic-9GGCN#111:1 Accuracy: 84.86 ± 4.55
node-classification-on-pubmed-48-32-20-fixedGGCN#151:1 Accuracy: 89.15 ± 0.37
node-classification-on-squirrelGGCN#24Accuracy: 55.17 ± 1.58
node-classification-on-texasGGCN#22Accuracy: 84.86 ± 4.55
node-classification-on-wisconsinGGCN#27Accuracy: 86.86 ± 3.29