Global-Local Graph Neural Networks for Node-Classification

Moshe Eliasof, Eran Treister

Abstract

The task of graph node classification is often approached blackby utilizing a local Graph Neural Network (GNN), that learns only local information from the node input features and their adjacency. In this paper, we propose to black improve the performance of node classification GNNs by utilizing both global and local information, specifically by learning label- and node- features. We therefore call our method Global-Local-GNN (GLGNN). To learn proper label features, for each label, we maximize the similarity between its features and nodes features that belong to the label, while maximizing the distance between nodes that do not belong to the considered label. We then use the learnt label features to predict the node classification map. We demonstrate our GLGNN using three different GNN backbones, and show that our approach improves baseline performance, revealing the importance of global information utilization blackfor node classification.

Benchmark Results

Tested on 5 datasets 28 baseline values reported too low

Tables show presented results in global context (best 3 papers + GCN + MLP).

TU Dortmund

MUTAG

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MSH-GNN GNN global	0.9910 ±0.0030	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
2	ID-GNN+BEC Hyb global	0.9830 ±0.0360	Depth-Adaptive Graph Neural Networks via Learnable Bakry-Émery Curvature
3	SupCosine Hyb global	0.9830 ±0.0250	Supervised Contrastive Learning with Structure Inference for Graph Classification
39	GIN GNN baseline actual #39 vs paper #272 233 places worse	0.9330 ±0.0290 best reported	actual #39 vs paper #272 233 places worse paper 0.8900 ±0.0600 best at pub: ProtGNN: Towards Self-Explaining Graph Neural Networks (AAAI 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.8900 ±0.0600 Δ at pub 0.0430
63	GCN GNN baseline actual #63 vs paper #432 369 places worse	0.9220 ±0.0440 best reported	actual #63 vs paper #432 369 places worse paper 0.8560 ±0.0580 best at pub: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) best today: Depth-Adaptive Graph Neural Networks via Learnable Bakry-Émery Curvature (KDD 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8560 ±0.0580 Δ at pub 0.0350 Δ today 0.0660
131	GLGIN Hyb baseline	0.9080 ±0.0610	This paper
158	GraphSAGE code ↗ GNN global	0.9040 ±0.0780	Asynchronous Neural Networks for Learning in Graphs code ↗
319	GLGCN Hyb baseline	0.8830 ±0.0510	This paper
403	MLP code ↗ Trad global	0.8660 ±0.0495	Breaking the Limits of Message Passing Graph Neural Networks code ↗

Heterophilic Graphs

Texas

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	P^2GNN Hyb global	0.9672 ±0.0147	$P^2$GNN: Two Prototype Sets to boost GNN Performance
2	DoG code ↗ ext data GNN global	0.9660 ±0.0210	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
3	GPF-plus code ↗ ext data GNN global	0.9583 ±0.0419	ProG: A Graph Prompt Learning Benchmark code ↗
28	GPRGNN GNN baseline actual #28 vs paper #185 157 places worse	0.9295 ±0.0131 best reported	actual #28 vs paper #185 157 places worse paper 0.8294 best at pub: Node-oriented Spectral Filtering for Graph Neural Networks (2022) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.8294 Δ at pub 0.0998 Δ today 0.1001
34	MLP Trad global	0.9230 ±0.0070	Design your own universe: a physics-informed agnostic method for enhancing graph neural networks
39	GGCN GNN baseline actual #39 vs paper #104 65 places worse	0.8686	actual #39 vs paper #104 65 places worse paper 0.8686 prev best: Understanding convolution on graphs via energies (TMLR 2022) best today: Graph Wave Networks (WWW 2025) Improvement within reported uncertainty; not highlighted. paper 0.8686 Δ at pub 0.0200 Δ today 0.0527
40	H2GCN GNN baseline actual #40 vs paper #94 54 places worse	0.8765	actual #40 vs paper #94 54 places worse paper 0.8765 prev best: LOGIN: A Large Language Model Consulted Graph Neural Network Training Framework (2024) paper 0.8765 Δ at pub 0.0098
46	APPNP GNN baseline actual #46 vs paper #438 392 places worse	0.9131 ±0.0197 best reported	actual #46 vs paper #438 392 places worse paper 0.6541 best at pub: A Quasi-Wasserstein Loss for Learning Graph Neural Networks (WWW 2023) paper 0.6541 Δ at pub 0.2590
53	GraphSAGE code ↗ GNN global	0.9000 ±0.0365	Graph Neural Networks Need Cluster-Normalize-Activate Modules code ↗
61	FAGCN GNN baseline actual #61 vs paper #185 124 places worse	0.8961 ±0.0152 best reported	actual #61 vs paper #185 124 places worse paper 0.8294 best at pub: Node-oriented Spectral Filtering for Graph Neural Networks (2022) paper 0.8294 Δ at pub 0.0667
70	GCNII* GNN baseline actual #70 vs paper #215 145 places worse	0.8157	actual #70 vs paper #215 145 places worse paper 0.8157 prev best: Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited (NeurIPS 2022) best today: Piecewise Constant Spectral Graph Neural Network (TMLR 2025) Improvement within reported uncertainty; not highlighted. paper 0.8157 Δ at pub 0.0111 Δ today 0.0761
81	GLGCNII Hyb baseline	0.8862	This paper
104	WRGAT GNN baseline actual #104 vs paper #104	0.8698	actual #104 vs paper #104 paper 0.8698 prev best: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) Improvement within reported uncertainty; not highlighted. paper 0.8698 Δ at pub 0.0336
211	GAT GNN baseline actual #211 vs paper #584 373 places worse	0.8162 ±0.0645 best reported	actual #211 vs paper #584 373 places worse paper 0.4941 best at pub: A Quasi-Wasserstein Loss for Learning Graph Neural Networks (WWW 2023) best today: Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models (2025) paper 0.4941 Δ at pub 0.3059 Δ today 0.3221
217	GRAND GNN baseline actual #217 vs paper #255 38 places worse	0.8140 ±0.0350 best reported	actual #217 vs paper #255 38 places worse paper 0.7941 best at pub: Unifying over-smoothing and over-squashing in graph neural networks: A physics informed approach and beyond (2023) best today: When Graph Neural Networks Meet Dynamic Mode Decomposition (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7941 Δ at pub 0.0079 Δ today 0.0199
257	GCN GNN baseline actual #257 vs paper #586 329 places worse	0.7933 ±0.0447 best reported	actual #257 vs paper #586 329 places worse paper 0.4892 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) best today: Graph Wave Networks (WWW 2025) paper 0.4892 Δ at pub 0.2948 Δ today 0.3041
259	PowerEmbed GNN baseline	0.7919	This paper
415	Geom-GCN GNN baseline actual #415 vs paper #449 34 places worse	0.6760 best reported	actual #415 vs paper #449 34 places worse paper 0.6412 best at pub: Non-Local Graph Neural Networks (2020) paper 0.6412 Δ at pub 0.0348
433	GLGAT Hyb baseline	0.6588	This paper
440	GLGCN Hyb baseline	0.6529	This paper
499	PairNorm GNN baseline actual #499 vs paper #591 92 places worse	0.6030 ±0.0430 best reported	actual #499 vs paper #591 92 places worse paper 0.4843 best at pub: Graph-Coupled Oscillator Networks (ICML 2022) best today: Dirichlet Energy Enhancement of Graph Neural Networks by Framelet Augmentation (2023) paper 0.4843 Δ at pub 0.1187 Δ today 0.1187

Cornell

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	P^2GNN Hyb global	0.9541 ±0.0272	$P^2$GNN: Two Prototype Sets to boost GNN Performance
2	ACM-GCN code ↗ GNN global	0.9480 ±0.0380	Heterophily-informed Message Passing code ↗
3	RF-GCN GNN global	0.9429	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
15	GGCN GNN baseline actual #15 vs paper #79 64 places worse	0.9230 ±0.0367 best reported	actual #15 vs paper #79 64 places worse paper 0.8568 best at pub: Understanding convolution on graphs via energies (TMLR 2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.8568 Δ today 0.0662
21	APPNP GNN baseline actual #21 vs paper #435 414 places worse	0.9181 ±0.0196 best reported	actual #21 vs paper #435 414 places worse paper 0.5430 best at pub: LON-GNN: Spectral GNNs with Learnable Orthonormal Basis (2023) paper 0.5430 Δ at pub 0.3751
22	GPRGNN GNN baseline actual #22 vs paper #177 155 places worse	0.9157 ±0.0196 best reported	actual #22 vs paper #177 155 places worse paper 0.8027 best at pub: Simplified PCNet with Robustness (2024) paper 0.8027 Δ at pub 0.1130
24	MLP Trad global	0.9136 ±0.0070	Node-oriented Spectral Filtering for Graph Neural Networks
32	GLGCNII Hyb baseline	0.8919	This paper
35	GCNII* GNN baseline actual #35 vs paper #218 183 places worse	0.8918 ±0.0396 best reported	actual #35 vs paper #218 183 places worse paper 0.7649 best at pub: Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited (NeurIPS 2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.7649 Δ at pub 0.0777 Δ today 0.1269
38	FAGCN GNN baseline actual #38 vs paper #185 147 places worse	0.8852 ±0.0133 best reported	actual #38 vs paper #185 147 places worse paper 0.7919 best at pub: Node-oriented Spectral Filtering for Graph Neural Networks (2022) paper 0.7919 Δ at pub 0.0933
68	H2GCN GNN baseline actual #68 vs paper #136 68 places worse	0.8623 ±0.0471 best reported	actual #68 vs paper #136 68 places worse paper 0.8270 best at pub: Design your own universe: a physics-informed agnostic method for enhancing graph neural networks (2024) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8270 Δ at pub 0.0240 Δ today 0.0353
138	SAGE GNN global	0.8269	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
142	GRAND GNN baseline actual #142 vs paper #151 9 places worse	0.8246 ±0.0709 best reported	actual #142 vs paper #151 9 places worse paper 0.8216 best at pub: Dirichlet Energy Enhancement of Graph Neural Networks by Framelet Augmentation (2023) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8216 Δ at pub 0.0004 Δ today 0.0030
158	GAT GNN baseline actual #158 vs paper #434 276 places worse	0.8190 ±0.0540 best reported	actual #158 vs paper #434 276 places worse paper 0.5432 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.5432 Δ at pub 0.2758
160	WRGAT GNN baseline	0.8162	This paper
170	GCN GNN baseline actual #170 vs paper #459 289 places worse	0.8082 ±0.0360 best reported	actual #170 vs paper #459 289 places worse paper 0.5270 best at pub: Differentiable Cluster Graph Neural Network (2024) best today: Bayesian Neighborhood Adaptation for Graph Neural Networks (TMLR 2025) paper 0.5270 Δ at pub 0.2745 Δ today 0.2812
195	PowerEmbed GNN baseline	0.7830	This paper
224	GLGAT Hyb baseline	0.7567	This paper
234	GLGCN Hyb baseline	0.7486	This paper
342	Geom-GCN GNN baseline actual #342 vs paper #369 27 places worse	0.6412 best reported	actual #342 vs paper #369 27 places worse paper 0.6081 best at pub: Steering Graph Neural Networks with Pinning Control (2023) paper 0.6081 Δ at pub 0.0331
389	PairNorm GNN baseline	0.5892	This paper

Actor

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DiGGR Hyb global	0.4535 ±0.0353	Disentangled Generative Graph Representation Learning
2	TRIGON code ↗ Hyb global	0.4402 ±0.0028	Dynamic Triangulation-Based Graph Rewiring for Graph Neural Networks code ↗
3	DoG code ↗ ext data GNN global	0.4320 ±0.0120	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
38	GCNII* GNN baseline actual #38 vs paper #278 240 places worse	0.4082 ±0.0179 best reported	actual #38 vs paper #278 240 places worse paper 0.3361 best at pub: Feature Expansion for Graph Neural Networks (ICML 2023) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3361 Δ at pub 0.0500 Δ today 0.0721
41	APPNP GNN baseline actual #41 vs paper #70 29 places worse	0.4046 ±0.0064 best reported	actual #41 vs paper #70 29 places worse paper 0.3886 best at pub: A Quasi-Wasserstein Loss for Learning Graph Neural Networks (WWW 2023) paper 0.3886 Δ at pub 0.0160
42	MLP Trad global	0.4026 ±0.0108	Heterophily-Aware Graph Attention Network
52	GPRGNN GNN baseline actual #52 vs paper #247 195 places worse	0.3992 ±0.0067 best reported	actual #52 vs paper #247 195 places worse paper 0.3463 best at pub: LON-GNN: Spectral GNNs with Learnable Orthonormal Basis (2023) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.3463 Δ at pub 0.0529 Δ today 0.0529
56	FAGCN GNN baseline actual #56 vs paper #240 184 places worse	0.3970 best reported	actual #56 vs paper #240 184 places worse paper 0.3487 best at pub: Beyond Low-frequency Information in Graph Convolutional Networks (AAAI 2021) paper 0.3487 Δ at pub 0.0483
62	GGCN GNN baseline actual #62 vs paper #99 37 places worse	0.3929 ±0.0048 best reported	actual #62 vs paper #99 37 places worse paper 0.3781 best at pub: Feature Transportation Improves Graph Neural Networks (AAAI 2023) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3781 Δ today 0.0148
68	GraphSAGE code ↗ GNN global	0.3899 ±0.0085	A Quasi-Wasserstein Loss for Learning Graph Neural Networks code ↗
71	H2GCN GNN baseline actual #71 vs paper #198 127 places worse	0.3885 ±0.0117 best reported	actual #71 vs paper #198 127 places worse paper 0.3570 best at pub: ES-GNN: Generalizing Graph Neural Networks Beyond Homophily with Edge Splitting (2022) best today: $P^2$GNN: Two Prototype Sets to boost GNN Performance (2026) paper 0.3570 Δ at pub 0.0270 Δ today 0.0315
75	WRGAT GNN baseline actual #75 vs paper #158 83 places worse	0.3860 ±0.0110 best reported	actual #75 vs paper #158 83 places worse paper 0.3653 best at pub: ES-GNN: Generalizing Graph Neural Networks Beyond Homophily with Edge Splitting (2022) paper 0.3653 Δ at pub 0.0207
87	Geom-GCN GNN baseline actual #87 vs paper #305 218 places worse	0.3810 best reported	actual #87 vs paper #305 218 places worse paper 0.3163 best at pub: Beyond Low-frequency Information in Graph Convolutional Networks (AAAI 2021) paper 0.3163 Δ at pub 0.0647
135	GCN GNN baseline actual #135 vs paper #408 273 places worse	0.3690 best reported	actual #135 vs paper #408 273 places worse paper 0.2732 best at pub: Beyond Low-frequency Information in Graph Convolutional Networks (AAAI 2021) paper 0.2732 Δ at pub 0.0958
182	GAT GNN baseline actual #182 vs paper #406 224 places worse	0.3605 ±0.0035 best reported	actual #182 vs paper #406 224 places worse paper 0.2744 best at pub: Hierarchical Graph Transformer with Adaptive Node Sampling (NeurIPS 2022) paper 0.2744 Δ at pub 0.0861
203	GRAND GNN baseline	0.3562	This paper
263	GLGCNII Hyb baseline	0.3413	This paper
296	GLGCN Hyb baseline	0.3245	This paper
305	GLGAT Hyb baseline	0.3184	This paper
408	PairNorm GNN baseline	0.2740	This paper

Wisconsin

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DMD++ Hyb global	0.9190 ±0.0260	When Graph Neural Networks Meet Dynamic Mode Decomposition
2	F-GREAD-BS code ↗ GNN global	0.9059 ±0.0380	Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND code ↗
3	APPNP GNN baseline actual #3 vs paper #114 111 places worse	0.9010 ±0.0350 best reported	actual #3 vs paper #114 111 places worse paper 0.7351 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) best today: When Graph Neural Networks Meet Dynamic Mode Decomposition (ICLR 2024) paper 0.7351 Δ at pub 0.1079 Δ today 0.1659
5	GPRGNN GNN baseline actual #5 vs paper #89 84 places worse	0.8990 ±0.0300 best reported	actual #5 vs paper #89 84 places worse paper 0.7838 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) best today: When Graph Neural Networks Meet Dynamic Mode Decomposition (ICLR 2024) paper 0.7838 Δ at pub 0.0712 Δ today 0.1152
6	GLGCNII Hyb baseline	0.8972	This paper
13	GRAND GNN baseline actual #13 vs paper #96 83 places worse	0.8860 ±0.0210 best reported	actual #13 vs paper #96 83 places worse paper 0.7568 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: When Graph Neural Networks Meet Dynamic Mode Decomposition (ICLR 2024) paper 0.7568 Δ at pub 0.0373 Δ today 0.1292
22	H2GCN GNN baseline actual #22 vs paper #45 23 places worse	0.8790 ±0.0420 best reported	actual #22 vs paper #45 23 places worse paper 0.8486 best at pub: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) best today: When Graph Neural Networks Meet Dynamic Mode Decomposition (ICLR 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8486 Δ at pub 0.0279 Δ today 0.0304
29	WRGAT GNN baseline actual #29 vs paper #58 29 places worse	0.8698 ±0.0378 best reported	actual #29 vs paper #58 29 places worse paper 0.8362 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8362 Δ at pub 0.0336 Δ today 0.0336
32	GGCN GNN baseline actual #32 vs paper #45 13 places worse	0.8686 ±0.0320 best reported	actual #32 vs paper #45 13 places worse paper 0.8486 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Sign is Not a Remedy: Multiset-to-Multiset Message Passing for Learning on Heterophilic Graphs (ICML 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8486 Δ at pub 0.0200 Δ today 0.0200
37	MLP code ↗ Trad global	0.8629 ±0.0450	Is Homophily a Necessity for Graph Neural Networks? code ↗
53	GAT GNN baseline actual #53 vs paper #172 119 places worse	0.8430 ±0.0370 best reported	actual #53 vs paper #172 119 places worse paper 0.5838 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.5838 Δ at pub 0.2592
60	FAGCN GNN baseline actual #60 vs paper #65 5 places worse	0.8333 best reported	actual #60 vs paper #65 5 places worse paper 0.8243 best at pub: Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2022) paper 0.8243 Δ at pub 0.0090
63	GCN GNN baseline actual #63 vs paper #202 139 places worse	0.8310 ±0.0320 best reported	actual #63 vs paper #202 139 places worse paper 0.5216 best at pub: Restructuring Graphs for Higher Homophily via Adaptive Spectral Clustering (AAAI 2022) paper 0.5216 Δ at pub 0.3094
69	GraphSAGE code ↗ GNN global	0.8118 ±0.0556	Graph Navier–Stokes Networks code ↗
75	GCNII* GNN baseline actual #75 vs paper #90 15 places worse	0.8039 ±0.0340 best reported	actual #75 vs paper #90 15 places worse paper 0.7784 best at pub: From Node Interaction to Hop Interaction: New Effective and Scalable Graph Learning Paradigm (2022) best today: Graph Navier–Stokes Networks (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.7784 Δ at pub 0.0255 Δ today 0.0255
89	PowerEmbed GNN baseline	0.7843	This paper
127	GLGCN Hyb baseline	0.7027	This paper
128	GLGAT Hyb baseline	0.7001	This paper
139	Geom-GCN GNN baseline actual #139 vs paper #139	0.6757	actual #139 vs paper #139 paper 0.6757 prev best: Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND (ICLR 2024) Improvement within reported uncertainty; not highlighted. paper 0.6757 Δ at pub 0.0306
169	PairNorm GNN baseline actual #169 vs paper #169	0.6027	actual #169 vs paper #169 paper 0.6027 prev best: Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-smoothness in Deep GNNs (AAAI 2023) paper 0.6027 Δ at pub 0.0743

SOTA performance (not significantly different from the best) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported result worse than the best known when published (best reported in green) Best at pub — reported value improved over the best known when published Best today — OK at publication; today's leaderboard best is elsewhere