LINKX Benchmarks
Deezer
Structure
1Graphs
28,281Nodes
1,650,703Edges
2Classes
Coverage
34Results
1Variants
3Papers
- ICLR 2026: Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors Learning Posterior Predictive Distributions for Node Classification from Synthetic Graph Priors ICLR 2026
- ICML 2025: Beyond Message Passing: Neural Graph Pattern Machine Beyond Message Passing: Neural Graph Pattern Machine ICML 2025
- TMLR 2022: Simplifying Node Classification on Heterophilous Graphs with Compatible Label Propagation Simplifying Node Classification on Heterophilous Graphs with Compatible Label Propagation TMLR 2022
Deezer Europe user friendship graph binary node classification (like/dislike), used in LINKX-style non-homophily evaluations.
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | ROC-AUC ↑ | Accuracy | Paper |
|---|---|---|---|---|
| 1 | GPM | 0.6726
±
0.0022 | — | |
| 2 | NodeFormer | 0.6640
±
0.0070 | — | |
| 3 | OrderedGNN | 0.6612
±
0.0075 | — | |
| 4 | APPNP | 0.6610
±
0.0060 | — | |
| 5 | NAGphormer | 0.6554
±
0.0057 | — | |
| 6 | GOAT | 0.6531
±
0.0024 | — | |
| 7 | VCR-Graphormer | 0.6528
±
0.0051 | — | |
| 8 | GCFormer | 0.6516
±
0.0033 | — | |
| 9 | RAW-GNN | 0.6511
±
0.0064 | — | |
| 10 | SAN | 0.6429
±
0.0035 | — | |
| 11 | RUM | 0.6425
±
0.0062 | — | |
| 12 | GPRGNN | 0.6320
±
0.0084 | — | |
| 13 | GCN | 0.6270
±
0.0070 | — | |
| 14 | GAT | 0.6170
±
0.0080 | — | |
| 15 | GraphGPS | 0.6056
±
0.0062 | — | |
| 16 | MLP | — Accuracy
0.6377
±
0.0030 | 0.6377
±
0.0030 | |
| 17 | LINK | — Accuracy
0.5595
±
0.0034 | 0.5595
±
0.0034 | |
| 18 | GCN | — Accuracy
0.5994
±
0.0055 | 0.5994
±
0.0055 | |
| 19 | GAT | — Accuracy
0.2434
±
0.0121 | 0.2434
±
0.0121 | |
| 20 | GCN2 | — Accuracy
0.6233
±
0.0081 | 0.6233
±
0.0081 | |
| 21 | MixHop | — Accuracy
0.6416
±
0.0085 | 0.6416
±
0.0085 | |
| 22 | SuperGAT | — Accuracy
0.5707
±
0.0064 | 0.5707
±
0.0064 | |
| 23 | GPRGNN | — Accuracy
0.6274
±
0.0039 | 0.6274
±
0.0039 | |
| 24 | FAGCN | — Accuracy
0.6504
±
0.0045 | 0.6504
±
0.0045 | |
| 25 | H2GCN | — Accuracy
0.6282
±
0.0068 | 0.6282
±
0.0068 | |
| 26 | CPGNN | — Accuracy
0.6495
±
0.0039 | 0.6495
±
0.0039 | |
| 27 | LP | — Accuracy
0.5544
±
0.0046 | 0.5544
±
0.0046 | |
| 28 | C&S | — Accuracy
0.6392
±
0.0071 | 0.6392
±
0.0071 | |
| 29 | CLP | — Accuracy
0.6569
±
0.0032 | 0.6569
±
0.0032 | |
| 30 | MLP | — Accuracy
0.5424
±
0.0215 | 0.5424
±
0.0215 | |
| 31 | GCN | — Accuracy
0.5369
±
0.0229 | 0.5369
±
0.0229 | |
| 32 | GAT | — Accuracy
0.5599
±
0.0378 | 0.5599
±
0.0378 | |
| 33 | GraphAny(Wisconsin) | — Accuracy
0.5213
±
0.0302 | 0.5213
±
0.0302 | |
| 34 | NodePFN | — Accuracy
0.5345
±
0.0065 | 0.5345
±
0.0065 |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time