Research map / Graph learning / GNN architectures
Plain Transformers are Surprisingly Powerful Link Predictors
Quang Truong, Yu Song, Donald Loveland, Mingxuan Ju, Tong Zhao, Neil Shah, Jiliang Tang
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 |
|---|---|---|---|
| add node features to PENCIL (ablation) | PENCIL · cora (original split) | MRR: 42.23 -> 32.12 | Hurt |
| add node features to PENCIL (ablation) | PENCIL · ogbl-ppa (original split) | Hits@100: 73.85 -> 79.54 | Improved |
| add node features to PENCIL (ablation) | PENCIL · ogbl-collab (original split) | Hits@50: 66.88 -> 66.56 | No clear effect |
| add node features to PENCIL (ablation) | PENCIL · ogbl-ppa (HeaRT) | MRR: 44.57 -> 45.43 | Improved |
| increase PENCIL depth (number of layers) (ablation) | PENCIL · cora, pubmed, ogbl-collab | Hits@50/Hits@100 | Improved |
| increase depth of standard GNNs for local heuristic estimation (ablation) | GNN/MPNN · cora heuristic regression (AA, CN, RA) | RMSE | No clear effect |
| add ID augmentation to GNNs for local heuristic estimation (ablation) | ID-GNN · cora heuristic regression | RMSE | Improved |
| use PENCIL instead of GNNs for PageRank heuristic estimation | PENCIL · cora heuristic regression (PageRank) | RMSE | Improved |
| use multiple labeled subgraphs with DeepSets-style pooling (ablation) | PENCIL · link prediction benchmarks | performance | No clear effect |
| use PENCIL instead of heuristic-informed baselines on ogbl-ppa | PENCIL · ogbl-ppa (original split) | Hits@100: 65.24 -> 79.54 | Improved |
| use PENCIL on small dataset citeseer vs LPFormer | PENCIL · citeseer (original split) | MRR: 65.42 -> 47.51 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MPLP+ — MPLP++: Fast, Parallel Dual Block-Coordinate Ascent for Dense Graphical Models (ECCV 2018) · better on 2 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.90)
- From Theory to Practice: Rethinking Green and Martin Kernels for Unleashing Graph Transformers (ICML 2025 · similarity 0.90)
- Structural Information Enhanced Graph Representation for Link Prediction (AAAI 2024 · similarity 0.90)
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.89)
- Bridging Theory and Practice in Link Representation with Graph Neural Networks (NeurIPS 2025 · similarity 0.89)
- Towards Principled Graph Transformers (NeurIPS 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.
