mlsanityData to 14 Oct 2026

When Design Rules Break: Benchmark Composition Determines Whether Label Informativeness Predicts GNN Aggregator Choice

arXiv 2606.10249 ↗First posted 8 Jun 2026cs.LG cs.SIcited by 0

Not on the map. Why?

Abstract (arXiv)

We examine whether graph neural network (GNN) design rules generalize across benchmark families by studying aggregator selection (sum, mean, max) on 24 node-classification datasets spanning citation, heterophilic, LINKX Facebook-100, co-purchase, and co-authorship graphs. Edge homophily is only weakly predictive of the GIN-Sum versus GIN-Mean performance gap. Label informativeness predicts this gap well on legacy benchmarks but degrades substantially when Facebook-100 graphs are included. In these dense friendship networks, near-zero label informativeness coexists with a strong preference for sum aggregation, producing gains of 7-10% and up to 13% under extended training. Stochastic block model ablations, including degree-corrected variants matching Facebook-100 degree scales, fail to reproduce this behavior, indicating that mean degree alone does not explain the effect. Among several label-independent graph statistics, the spectral gap uniquely distinguishes these graphs from other low-informativeness datasets, with the effect localized to one-hop neighborhoods and replicated across architectures. We further identify training regimes that interact with aggregator choice and show that PNA can underperform the best single-aggregator GIN on standard citation benchmarks. Our results suggest that benchmark composition, rather than numerical insufficiency, determines whether design rules appear to generalize, and that the Facebook-100 regime provides a concrete target for future adaptive aggregation methods.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 21 of its 24 references that are in the map. First posted on arXiv 8 Jun 2026.

Builds on

21 of 24 references.

References · 21

  1. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,361
  2. 2017Graph Attention Networkscited by 8,923
  3. 2017Inductive Representation Learning on Large Graphscited by 6,320
  4. 2018How Powerful are Graph Neural Networks?cited by 4,393
  5. 2020Open Graph Benchmark: Datasets for Machine Learning on Graphscited by 2,163
  6. 2018Pitfalls of Graph Neural Network Evaluationcited by 1,007
  7. 2016Revisiting Semi-Supervised Learning with Graph Embeddingscited by 976
  8. 2020Geom-GCN: Geometric Graph Convolutional Networkscited by 826
  9. 2020Adaptive Universal Generalized PageRank Graph Neural Networkcited by 595
  10. 2019Cluster-GCN: An Efficient Algorithm for Training Deep and Large Graph Convolutional Networkscited by 541
  11. 2020Principal Neighbourhood Aggregation for Graph Netscited by 478
  12. 2019MixHop: Higher-Order Graph Convolutional Architectures via Sparsified Neighborhood Mixingcited by 443
9 more
  1. 2023A critical look at the evaluation of GNNs under heterophily: Are we really making progress?cited by 343
  2. 2021Large Scale Learning on Non-Homophilous Graphs: New Benchmarks and Strong Simple Methodscited by 311
  3. 2021Is Homophily a Necessity for Graph Neural Networks?cited by 182
  4. 2022Revisiting Heterophily For Graph Neural Networkscited by 169
  5. 2024Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classificationcited by 85
  6. 2021Simplifying approach to Node Classification in Graph Neural Networkscited by 56
  7. 2023When Do Graph Neural Networks Help with Node Classification? Investigating the Impact of Homophily Principle on Node Distinguishabilitycited by 7
  8. 2025Beyond Fixed Depth: Adaptive Graph Neural Networks for Node Classification Under Varying Homophilycited by 5
  9. 2024Revisiting Neighborhood Aggregation in Graph Neural Networks for Node Classification using Statistical Signal Processingcited by 2

Built on by

Not cited yet.