Research map / Graph learning / GNN architectures
Is Fixing Schema Graphs Necessary? Full-Resolution Graph Structure Learning for Relational Deep Learning
Yi Huang, Qingyun Sun, Jia Li, Xingcheng Fu, Jianxin Li
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 |
|---|---|---|---|
| learn table roles (node vs edge) jointly with the GNN (FROG) instead of fixed schema graph | RDL GraphSAGE / RelGNN · rel-event user-ignore | ROC-AUC: 82.55 -> 85.49 (vs RelGNN) | Improved |
| learn table roles instead of fixed roles on rel-hm item-sales regression | RDL GraphSAGE · rel-hm item-sales | MAE: 0.0556 (RDL) vs 0.0561 (FROG) | No clear effect |
| learn table roles (FROG) on rel-trial study-adverse | LightGBM · rel-trial study-adverse | MAE: 42.8452 (LightGBM) vs 44.7536 (FROG) | Hurt |
| learn table roles on site-sponsor-run recommendation | RelGNN · rel-trial site-sponsor-run | MAP: 19.10 (RelGNN) vs 19.09 (FROG) | No clear effect |
| learned node/edge role mixing versus all-node, all-edge, or random weights (ablation) | FROG table roles · rel-event user-repeat, rel-avito ad-ctr, rel-stack post-post-related | ROC-AUC / MAE / MAP | Improved |
| model all tables as nodes vs all as edges (ablation) | FROG variants · rel-event user-repeat vs rel-avito ad-ctr | ROC-AUC / MAE | Mixed |
| add functional dependency losses L_emb and L_pair (ablation) | FROG representations · rel-event, rel-avito (embedding visualization) | embedding separability (qualitative) | Improved |
| transfer structure learned from other tasks (rel-event user-repeat) (ablation) | FROG · rel-event user-repeat | ROC-AUC: 79.01 -> 76.30 / 74.31 | Hurt |
| transfer structure learned from other tasks (rel-event user-attendance) (ablation) | FROG · rel-event user-attendance | MAE: 0.2435 vs 0.2635 | Improved |
| transfer structures across rel-avito tasks (ablation) | FROG · rel-avito | ROC-AUC / MAE / MAP: user-visits 66.37 vs 66.54 transferred | No clear effect |
Closest papers by meaning
- Position: Relational Deep Learning - Graph Representation Learning on Relational Databases (ICML 2024 · similarity 0.92)
- RelGNN: Composite Message Passing for Relational Deep Learning (ICML 2025 · similarity 0.89)
- RDB2G-Bench: A Comprehensive Benchmark for Automatic Graph Modeling of Relational Databases (NeurIPS 2025 · similarity 0.87)
- Joint Relational Database Generation via Graph-Conditional Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Griffin: Towards a Graph-Centric Relational Database Foundation Model (ICML 2025 · similarity 0.86)
- 4DBInfer: A 4D Benchmarking Toolbox for Graph-Centric Predictive Modeling on RDBs (NeurIPS 2024 · similarity 0.86)
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.
