Pith. sign in

REVIEW 10 cited by

A critical look at the evaluation of GNNs under heterophily: Are we really making progress?

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2302.11640 v2 pith:RJLGOLEU submitted 2023-02-22 cs.LG

Oleg Platonov , Denis Kuznedelev , Michael Diskin , Artem Babenko , Liudmila Prokhorenkova This is my paper
classification cs.LG
keywords graphsgnnsdatasetsheterophilousnodesperformanceresultsstandard
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Node classification is a classical graph machine learning task on which Graph Neural Networks (GNNs) have recently achieved strong results. However, it is often believed that standard GNNs only work well for homophilous graphs, i.e., graphs where edges tend to connect nodes of the same class. Graphs without this property are called heterophilous, and it is typically assumed that specialized methods are required to achieve strong performance on such graphs. In this work, we challenge this assumption. First, we show that the standard datasets used for evaluating heterophily-specific models have serious drawbacks, making results obtained by using them unreliable. The most significant of these drawbacks is the presence of a large number of duplicate nodes in the datasets Squirrel and Chameleon, which leads to train-test data leakage. We show that removing duplicate nodes strongly affects GNN performance on these datasets. Then, we propose a set of heterophilous graphs of varying properties that we believe can serve as a better benchmark for evaluating the performance of GNNs under heterophily. We show that standard GNNs achieve strong results on these heterophilous graphs, almost always outperforming specialized models. Our datasets and the code for reproducing our experiments are available at https://github.com/yandex-research/heterophilous-graphs

Discussion (0). Sign in to comment.

Forward citations

Cited by 10 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. L2G-Net: Local to Global Spectral Graph Neural Networks via Cauchy Factorizations

    cs.LG 2026-02 conditional novelty 7.0 of 10

    The graph Fourier transform is exactly factored into a chain of local subgraph transforms stitched by Cauchy matrices, giving L2G-Net a spectral GNN with O(kn^2) setup cost and competitive long-range benchmarks.

  2. HeAD-CP: Heterophily-Aware Diffused Conformal Prediction Sets for Graph Neural Networks

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Graph conformal prediction with node-wise softmax-derived diffusion coefficients avoids DAPS's heterophily failure and beats it on 8/10 benchmarks in oracle selection.

  3. Towards Text-free Graph Foundation Models: Rethinking Multi-Domain Graph Contrastive Learning

    cs.CL 2025-06 conditional novelty 6.0 of 10

    MDGCL pre-trains graph encoders on multiple domains using same-domain discrimination and a downstream domain-attention mechanism, outperforming existing text-free graph foundation models.

  4. How to Use Graph Data in the Wild to Help Graph Anomaly Detection?

    cs.LG 2025-06 conditional novelty 6.0 of 10

    Wild-GAD selects relevant and diverse external graphs via a target-trained model and trains the detector on them, reporting large accuracy gains over baselines.

  5. AnomalyGFM: Graph Foundation Model for Zero/Few-shot Anomaly Detection

    cs.LG 2025-02 conditional novelty 6.0 of 10

    AnomalyGFM aligns learned normal and abnormal prototypes with node-neighbor residual features, enabling zero-shot and few-shot graph anomaly detection across datasets.

  6. SGS-GNN: A Supervised Graph Sparsification method for Graph Neural Networks

    cs.LG 2025-02 conditional novelty 5.0 of 10

    A supervised edge-sampling sparsifier that keeps only 20% of edges can match or beat the full graph for GNN node classification, with especially large gains on heterophilic graphs.

  7. Flow Matters: Directional and Expressive GNNs for Heterophilic Graphs

    cs.LG 2025-08 reject novelty 4.0 of 10

    A GAT model with polynomial gating (Poly) and a directed variant (Dir-Poly) report strong heterophilic node classification results, with Dir-Poly's largest gain on a single directed dataset.

  8. Intellectual Property in Graph-Based Machine Learning as a Service: Attacks and Defenses

    cs.CR 2025-08 conditional novelty 4.0 of 10

    A systematic review that organizes graph-ML IP protection into model-level and data-level attacks and defenses, and ships a benchmark library, PyGIP.

  9. Molecular Machine Learning Using Euler Characteristic Transforms

    cs.LG 2025-07 conditional novelty 4.0 of 10

    An ECT-based topological representation, combined with the AVALON fingerprint, ranks among the best-performing methods on most of nine Ki prediction benchmarks.

  10. S2FGL: Spatial Spectral Federated Graph Learning

    cs.LG 2025-07 conditional novelty 4.0 of 10

    S2FGL improves subgraph federated graph learning by injecting prototype-based semantic knowledge and aligning local and global spectral projections, gaining about 1 to 2 points of node classification accuracy.

Pith tools