Research map / Graph learning / Knowledge and temporal graphs
A Theory of Link Prediction via Relational Weisfeiler-Leman on Knowledge Graphs
Xingyue Huang, Miguel Romero Orth, Ismail Ilkan Ceylan, Pablo Barcelo
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 |
|---|---|---|---|
| Switch history from f(t)=0 to f(t)=t (ablation) | C-MPNN architectures · WN18RR and FB15k-237 inductive splits | Hits@10 | No clear effect |
| Use Msg2 instead of Msg1 or Msg3 (ablation) | C-MPNN · FB15k-237 inductive splits | Hits@10 | Hurt |
| Use Msg2 instead of other message functions (ablation) | C-MPNN · WN18RR inductive splits | Hits@10 | No clear effect |
| Use PNA rather than sum aggregation (ablation) | C-MPNN with Msg1 · FB15k-237 inductive splits | Hits@10 | Improved |
| Use PNA rather than sum aggregation (ablation) | C-MPNN with Msg2 or Msg3 · FB15k-237 inductive splits | Hits@10 | Hurt |
| Use PNA rather than sum aggregation (ablation) | C-MPNN · WN18RR inductive splits | Hits@10 | Mixed |
| Use an initialization that fails target node distinguishability (ablation) | C-MPNN · Inductive link prediction | Hits@10 | Hurt |
| Add a global readout (ablation) | C-MPNN · WN18RR inductive splits | Hits@10 | No clear effect |
| Add a global readout (ablation) | C-MPNN · FB15k-237 v2 and v4 | Hits@10: 0.949 -> 0.897; 0.954 -> 0.864 | Hurt |
| Use relation-specific readout instead of no readout (ablation) | C-MPNN · FB15k-237 v1-v4 | Hits@10: 0.804 -> 0.835; 0.949 -> 0.959; 0.945 -> 0.953; 0.954 -> 0.960 | Improved |
| Use relation-specific readout instead of no readout (ablation) | C-MPNN · WN18RR v1-v4 | Hits@10: 0.941 -> 0.932; 0.895 -> 0.885; 0.904 -> 0.882; 0.886 -> 0.874 | Hurt |
| Condition representations on a source node to produce pairwise representations | C-MPNN versus R-MPNN · Transductive link prediction and biomedical knowledge | Hits@10 | Improved |
Closest papers by meaning
- Bridging Theory and Practice in Link Representation with Graph Neural Networks (NeurIPS 2025 · similarity 0.89)
- Toward a Unified Statistical Theory of Unsupervised Pretraining and Supervised Neural Knowledge Graph Learning (NeurIPS 2026 · similarity 0.88)
- Knowledge Graphs Can be Learned with Just Intersection Features (ICML 2024 · similarity 0.88)
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.87)
- Learning Latent Graph Structures and their Uncertainty (ICML 2025 · similarity 0.87)
- Talk like a Graph: Encoding Graphs for Large Language Models (ICLR 2024 · 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.
