Heterophily Benchmark
Roman-empire
Structure
1Graphs
22,662Nodes
32,927Edges
18Classes
300Node Feature Dim
0.05Homophily
Coverage
308Results
1Variants
40Papers
- Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine Learning Graph Cascades: Contagion-Based Mesoscopic Rewiring for Structure-Aware Graph Machine Learning
- Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification
- Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
- Geometric Evolution Graph Convolutional Networks: Enhancing Graph Representation Learning via Ricci Flow Geometric Evolution Graph Convolutional Networks: Enhancing Graph Representation Learning via Ricci Flow
- Position: Don't be Afraid of Over-Smoothing And Over-Squashing Position: Don't be Afraid of Over-Smoothing And Over-Squashing
- ATLAS: Adaptive Topology-based Learning at Scale for Homophilic and Heterophilic Graphs ATLAS: Adaptive Topology-based Learning at Scale for Homophilic and Heterophilic Graphs
- Torch Geometric Pool: the Pytorch library for pooling in Graph Neural Networks Torch Geometric Pool: the Pytorch library for pooling in Graph Neural Networks
- ICML 2025: View Space: Learning Representation across Arbitrary Graphs View Space: Learning Representation across Arbitrary Graphs ICML 2025
- Adaptive Graph Mixture of Residual Experts: Unsupervised Learning on Diverse Graphs with Heterogeneous Specialization Adaptive Graph Mixture of Residual Experts: Unsupervised Learning on Diverse Graphs with Heterogeneous Specialization
- NeurIPS 2025: Relieving the Over-Aggregating Effect in Graph Transformers Relieving the Over-Aggregating Effect in Graph Transformers NeurIPS 2025
- Attention Beyond Neighborhoods: Reviving Transformer for Graph Clustering Attention Beyond Neighborhoods: Reviving Transformer for Graph Clustering
- Dynamic Triangulation-Based Graph Rewiring for Graph Neural Networks Dynamic Triangulation-Based Graph Rewiring for Graph Neural Networks
- Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax
- Aligning the Spectrum: Hybrid Graph Pre-training and Prompt Tuning across Homophily and Heterophily Aligning the Spectrum: Hybrid Graph Pre-training and Prompt Tuning across Homophily and Heterophily
- TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
- Aggregation-aware MLP: An Unsupervised Approach for Graph Message-passing Aggregation-aware MLP: An Unsupervised Approach for Graph Message-passing
- Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models
- Directed Homophily-Aware Graph Neural Network Directed Homophily-Aware Graph Neural Network
- Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding
- TMLR 2025: Heterophily-informed Message Passing Heterophily-informed Message Passing TMLR 2025
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning
- Scale-aware Message Passing For Graph Node Classification Scale-aware Message Passing For Graph Node Classification
- NeurIPS 2024: Even Sparser Graph Transformers Even Sparser Graph Transformers NeurIPS 2024
- NeurIPS 2024: Distributed-Order Fractional Graph Operating Network Distributed-Order Fractional Graph Operating Network NeurIPS 2024
- ICML 2024: Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms Diss-l-ECT: Dissecting Graph Data with local Euler Characteristic Transforms ICML 2024
- Re-evaluating the Advancements of Heterophilic Graph Learning Re-evaluating the Advancements of Heterophilic Graph Learning
- HyperAggregation: Aggregating over Graph Edges with Hypernetworks HyperAggregation: Aggregating over Graph Edges with Hypernetworks
- NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification NTFormer: A Composite Node Tokenized Graph Transformer for Node Classification
- Differentiable Cluster Graph Neural Network Differentiable Cluster Graph Neural Network
- Bundle Neural Networks for message diffusion on graphs Bundle Neural Networks for message diffusion on graphs
- ICLR 2024: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND ICLR 2024
- TMLR 2024: Contextualized Messages Boost Graph Representations Contextualized Messages Boost Graph Representations TMLR 2024
- Provable Filter for Real-world Graph Clustering Provable Filter for Real-world Graph Clustering
- ICLR 2024: Polynormer: Polynomial-Expressive Graph Transformer in Linear Time Polynormer: Polynomial-Expressive Graph Transformer in Linear Time ICLR 2024
- KDD 2024: Graph Mamba: Towards Learning on Graphs with State Space Models Graph Mamba: Towards Learning on Graphs with State Space Models KDD 2024
- CAT: A Causal Graph Attention Network for Trimming Heterophilic Graphs CAT: A Causal Graph Attention Network for Trimming Heterophilic Graphs
- ICLR 2023: HoloNets: Spectral Convolutions do extend to Directed Graphs HoloNets: Spectral Convolutions do extend to Directed Graphs ICLR 2023
- ICML 2023: Evolving Computation Graphs Evolving Computation Graphs ICML 2023
- IJCAI 2023: Graph Neural Convection-Diffusion with Heterophily Graph Neural Convection-Diffusion with Heterophily IJCAI 2023
- Graph Neural Networks for Graphs with Heterophily: A Survey Graph Neural Networks for Graphs with Heterophily: A Survey
18-class node classification Wikipedia Roman Empire article node classification (18 syntactic roles) from the heterophilic benchmark. Uses the corrected benchmark features and ten official splits. Evaluated by Accuracy.
Dataset details
Task
Wikipedia Roman Empire article node classification (18 syntactic roles) from the heterophilic benchmark. The primary catalog metric is Accuracy.
Data
Wikipedia Roman Empire article node classification (18 syntactic roles) from the heterophilic benchmark.
Features
The benchmark provides one fixed graph, node features, labels, and ten official splits per dataset. Features are dataset-specific and include text-derived, product, synthetic-grid, or platform-account attributes described by the release.
Splits and evaluation
The catalog records these protocols or variants: Standard split. Evaluation uses Accuracy (higher is better). Exact masks or folds must come from the cited release.
Quirks and pitfalls
Use the corrected release and its ten masks. Several tasks are imbalanced and use ROC-AUC rather than accuracy; treating missing relationships as negative evidence or tuning on the test masks invalidates comparison.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | AP | AUROC | ROC-AUC | macro-F1 | Paper |
|---|---|---|---|---|---|---|---|
| 1 | Dir-Poly | 0.9451
±
0.0022 | — | — | — | — | |
| 2 | D-CDE | 0.9387
±
0.0041 | — | — | — | — | |
| 3 | LargeScaleNet | 0.9358
±
0.0024 | — | — | — | — | |
| 4 | SSN | 0.9352
±
0.0028 | — | — | — | — | |
| 5 | GraphTARIF | 0.9323
±
0.0038 | — | — | — | — | |
| 6 | F-CDE | 0.9306
±
0.0055 | — | — | — | — | |
| 7 | Polynormer-r | 0.9255
±
0.0037 | — | — | — | — | |
| 8 | GHC | 0.9227
±
0.0057 | — | — | — | — | |
| 9 | FaberNet | 0.9224
±
0.0043 | — | — | — | — | |
| 10 | CoED | 0.9217
±
0.0029 | — | — | — | — | |
| 11 | Polynormer | 0.9213
±
0.0050 | — | — | — | — | |
| 12 | Polynormer-v1 | 0.9213
±
0.0050 | — | — | — | — | |
| 13 | Tango -GatedGCN | 0.9189
±
0.0030 | — | — | — | — | |
| 14 | NDDGNN | 0.9176
±
0.0027 | — | — | — | — | |
| 15 | BuNN | 0.9175
±
0.0039 | — | — | — | — | |
| 16 | GCN+TS | 0.9166
±
0.0020 | — | — | — | — | |
| 17 | CDE-GRAND-GAT | 0.9164
±
0.0028 | — | — | — | — | |
| 18 | CDE | 0.9164
±
0.0028 | — | — | — | — | |
| 19 | CDE-GRAND | 0.9164
±
0.0028 | — | — | — | — | |
| 20 | CDE-GRAND-TRANS | 0.9155
±
0.0023 | — | — | — | — | |
| 21 | Polynormer-v2 | 0.9133
±
0.0039 | — | — | — | — | |
| 22 | DirGNN | 0.9130
±
0.0046 | — | — | — | — | |
| 23 | SAGE+T | 0.9127
±
0.0036 | — | — | — | — | |
| 24 | Dir-GNN | 0.9123
±
0.0032 | — | — | — | — | |
| 25 | Tango -GPS | 0.9108
±
0.0057 | — | — | — | — | |
| 26 | GraphSAGE | 0.9106
±
0.0027 | — | — | — | — | |
| 27 | GAT+T | 0.9093
±
0.0020 | — | — | — | — | |
| 28 | CDE-GRAND-LAP | 0.9058
±
0.0049 | — | — | — | — | |
| 29 | Polynormer | 0.9032
±
0.0036 | — | — | — | — | |
| 30 | DC-GNN | 0.8996
±
0.0035 | — | — | — | — | |
| 31 | ECG-GAT-sep | 0.8962
±
0.0018 | — | — | — | — | |
| 32 | BGRL-ECG-GAT-sep | 0.8962
±
0.0018 | — | — | — | — | |
| 33 | ECG-GT-sep | 0.8956
±
0.0016 | — | — | — | — | |
| 34 | BGRL-ECG-GT-sep | 0.8956
±
0.0016 | — | — | — | — | |
| 35 | AGCN | 0.8940 | — | — | — | — | |
| 36 | PD-GNN | 0.8923
±
0.0056 | — | — | — | — | |
| 37 | DHGNN | 0.8905
±
0.0053 | — | — | — | — | |
| 38 | Exphormer | 0.8903
±
0.0037 | — | — | — | — | |
| 39 | Exphormer+GAT | 0.8903
±
0.0037 | — | — | — | — | |
| 40 | MPSNN | 0.8876
±
0.0063 | — | — | — | — | |
| 41 | GAT-sep | 0.8875
±
0.0041 | — | — | — | — | |
| 42 | MLPBGRL-GAT-sep | 0.8873
±
0.0037 | — | — | — | — | |
| 43 | MLPBGRL-GT-sep | 0.8870
±
0.0030 | — | — | — | — | |
| 44 | MLP->GNN-ECG-GT-sep | 0.8862
±
0.0046 | — | — | — | — | |
| 45 | MLP-ECG-GT-sep | 0.8856
±
0.0035 | — | — | — | — | |
| 46 | GCN-SSM (Optimal L) | 0.8837
±
0.0060 | — | — | — | — | |
| 47 | MLP-ECG-GAT-sep | 0.8822
±
0.0036 | — | — | — | — | |
| 48 | MagNet | 0.8807
±
0.0027 | — | — | — | — | |
| 49 | MLP->GNN-ECG-GAT-sep | 0.8804
±
0.0032 | — | — | — | — | |
| 50 | ECG-SAGE | 0.8788
±
0.0025 | — | — | — | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time