Research map / Graph learning / GNN architectures
GSLB: The Graph Structure Learning Benchmark
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Learn a task-adapted graph structure instead of using the original graph | GSL node-classification methods · Homophilic and heterophilic node-classification datasets, TR | accuracy: Cora: GCN 81.46{ $\pm$0.58} -> IDGL 83.88{ $\pm$0.42}; Texas: GPRGNN 85.14{ $\pm$3.68} vs SUBLIME 77.03{ $\pm$4.23} | Mixed |
| Apply GSL in topology-inference settings using feature-derived initial graphs where needed | GSL node-classification methods · Heterophilic node-classification datasets, TI | accuracy: Texas: GPRGNN kNN 81.08{ $\pm$4.18}; SLAPS 79.19{ $\pm$4.99} | Mixed |
| Model heterogeneous interactions and structure jointly | GTN and HGSL · ACM, DBLP, and Yelp | Macro-F1: ACM: HAN 91.67{ $\pm$0.39} -> HGSL 93.23{ $\pm$0.50}; DBLP: HAN 90.53{ $\pm$0.24} -> HGSL 91.58{ $\pm$0.40}; Yelp: HAN 8 | Helped |
| Refine graph structures for graph classification | HGP-SL · Graph-level classification datasets | accuracy: IMDB-B: GCN 73.20{ $\pm$4.29} -> HGP-SL (GCN) 74.10{ $\pm$4.55}; MUTAG: GCN 73.92{ $\pm$8.84} -> HGP-SL (GCN) 78.07{ $\p | Mixed |
| Use graph structure learning on long-range graph tasks | HGP-SL and VIB-GSL · Peptides-Func | AP: GIN 60.92{ $\pm$0.88} vs HGP-SL (GIN) 55.30{ $\pm$0.88} | Hurt |
| Use graph structure learning on long-range graph tasks | HGP-SL and VIB-GSL · Peptides-Struct | MAE: GIN 0.3647{ $\pm$0.0028} -> HGP-SL (GIN) 0.2786{ $\pm$0.0018} | Helped |
| Use self-supervised structure learning | GSL models under scarce supervision · Cora and Citeseer | accuracy | Helped |
| Use GSL rather than GCN under graph topology noise | GSL methods · Cora, Citeseer, and Polblogs | accuracy: Polblogs, 20%: GCN 68.19{ $\pm$2.24}; SUBLIME 92.60{ $\pm$0.72} | Helped |
| Increase graph-structure learning capacity | GSL models compared with GCN · Cora, Citeseer, and Pubmed | time and space complexity | Hurt |
| Use unsupervised GSL for robust structure modeling | STABLE and SUBLIME · Corrupted graph structures | accuracy: Cora, 20%: GCN 61.98{ $\pm$1.23}; STABLE 76.44{ $\pm$2.47}; SUBLIME 75.25{ $\pm$1.08} | Helped |
Closest papers by meaning
- GLBench: A Comprehensive Benchmark for Graph with Large Language Models (NeurIPS 2024 · similarity 0.89)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.89)
- GC-Bench: An Open and Unified Benchmark for Graph Condensation (NeurIPS 2024 · similarity 0.88)
- Structural Entropy Based Graph Structure Learning for Node Classification (AAAI 2024 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- No Metric to Rule Them All: Toward Principled Evaluations of Graph-Learning Datasets (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.