Research map / Graph learning / GNN architectures
Neural Link Prediction with Walk Pooling
Liming Pan, Cheng Shi, Ivan Dokmanić
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use WalkPool with distance labels instead of WalkPool with all-ones input features | WalkPool · USAir, NS, PB, Yeast, C.ele, Power, Router, E.coli | AUC: 98.52±0.50 -> 98.68±0.48; 98.86±0.42 -> 98.95±0.41; 95.42±0.39 -> 95.60±0.37; 98.16±0.33 -> 98.37±0.25; 92.42±1.22 -> 92 | Improved |
| Add WalkPool to unsupervised GNN representations | VGAE, ARGVA, and GIC · Cora, Citeseer, Pubmed, Chameleon, Cornell, Texas, Wisconsin | AUC: 91.98±0.54 -> 94.64±0.55; 91.21±1.14 -> 94.32±0.90; 96.51±0.14 -> 98.49±0.13; 98.79±0.18 -> 99.51±0.08; 70.59±9.03 -> 78 | Improved |
| Remove the ω₁₂ feature (ablation) | WalkPool · C.ele | AUC: 92.82 -> 92.69 | Hurt |
| Remove node-walk features (ablation) | WalkPool · C.ele | AUC: 92.82 -> 92.66 | Hurt |
| Remove link-walk features (ablation) | WalkPool · C.ele | AUC: 92.82 -> 91.08 | Hurt |
| Remove graph-walk features (ablation) | WalkPool · C.ele | AUC: 92.82 -> 92.65 | Hurt |
| Use the final-layer embedding only instead of concatenating initial and GCN-layer embeddings (ablation) | WalkPool · Link prediction datasets with node attributes | AUC | Hurt |
| Increase the walk-length cutoff beyond τc = 7 (ablation) | WalkPool · Link prediction benchmarks | accuracy | No clear effect |
| Use BCE loss instead of MSE loss (ablation) | WalkPool · Eight datasets without node attributes | AUC: 98.68/98.68; 98.95/98.85; 95.60/95.69; 98.37/98.37; 92.79/92.83; 92.56/92.58; 97.27/97.35; 98.58/98.67 | No clear effect |
Closest papers by meaning
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.91)
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.90)
- Weisfeiler and Lehman Go Paths: Learning Topological Features via Path Complexes (AAAI 2024 · similarity 0.89)
- Bridging Theory and Practice in Link Representation with Graph Neural Networks (NeurIPS 2025 · similarity 0.88)
- MaxCutPool: differentiable feature-aware Maxcut for pooling in graph neural networks (ICLR 2025 · similarity 0.87)
- Geometry-Aware Edge Pooling for Graph Neural Networks (NeurIPS 2025 · similarity 0.87)
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.
