Node Classification 벤치마크
Node Classification on Pubmed Full-supervised
Accuracy
- 2017-06-07 — GraphSAGE: Accuracy 87.1
- 2018-01-30 — FastGCN: Accuracy 88.0
- 2018-09-14 — ASGCN: Accuracy 90.6
- 2019-07-25 — GraphSAGE+DropEdge: Accuracy 91.7
| Rank | Model | Accuracy | Paper | Code | Year |
|---|---|---|---|---|---|
| 1 | GraphSAGE+DropEdge | 91.70% | DropEdge: Towards Deep Graph Convolutional Networks on Node Classification | GraphSAINT/GraphSAINT · GraphSAINT/GraphSAINT · DropEdge/DropEdge · +4 | 2019 |
| 2 | ASGCN | 90.6% | Adaptive Sampling Towards Fast Graph Representation Learning | dmlc/dgl · huangwb/AS-GCN | 2018 |
| 3 | FDGATII | 90.3524% | FDGATII : Fast Dynamic Graph Attention with Initial Residual and Identity Mapping | gayanku/FDGATII | 2021 |
| 4 | GCNII* | 90.30% | Simple and Deep Graph Convolutional Networks | chennnM/GCNII · chennnM/GCNII · zhanglab-aim/cancer-net · +1 | 2020 |
| 5 | Graph ESN | 89.2±0.3 | Beyond Homophily with Graph Echo State Networks | 2022 | |
| 6 | FastGCN | 88.00% | FastGCN: Fast Learning with Graph Convolutional Networks via Importance Sampling | matenure/FastGCN · gkunnan97/fastgcn_pytorch · jiechenjiechen/FastGCN-matlab · +1 | 2018 |
| 7 | GraphSAGE | 87.1% | Inductive Representation Learning on Large Graphs | pyg-team/pytorch_geometric · dmlc/dgl · dmlc/dgl · +17 | 2017 |