Research map / Graph learning / GNN architectures
Graph Neural Networks for Link Prediction with Subgraph Sketching
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 |
|---|---|---|---|
| Add sketch-based structure features (ELPH) instead of explicit subgraph construction | full-graph MPNN for link prediction · Cora / Citeseer / Pubmed / Collab / DDI | HR@100 / HR@50 / HR@20 | Helped |
| Precompute sketches and node features (BUDDY) instead of full-graph GNN propagation | ELPH · Collab / PPA / Citation2 / DDI | HR@50 / HR@100 / MRR / HR@20 | Mixed |
| Replace GNN structure-feature propagation with an MLP on DE structure-feature counts (max distance 3) (ablation) | SGNN structure features · three datasets | test performance | Mixed |
| Use edge (link-level) readout instead of mean or sum node pooling over subgraph (ablation) | SGNN readout · Planetoid / analysis datasets | predictive performance | Helped |
| Pre-propagate node features by neighbor averaging rather than raw features in an MLP with structure counts (ablation) | MLP on structure counts and node features · three datasets | test performance | Helped |
| Add structure features (DRNL/DE labels) to subgraph models (ablation) | SGNN · Planetoid datasets | test performance | Helped |
| Use BUDDY instead of SEAL dynamic subgraph generation | SEAL · Citation dataset training and inference | wall time: 200-1000x faster than SEAL dynamic | Helped |
Closest papers by meaning
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.92)
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.90)
- Mixture of Link Predictors on Graphs (NeurIPS 2024 · similarity 0.90)
- Bridging Theory and Practice in Link Representation with Graph Neural Networks (NeurIPS 2025 · similarity 0.90)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · similarity 0.89)
- Revisiting Link Prediction: a data perspective (ICLR 2024 · similarity 0.89)
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.