Gaussian Relational Graph Transformer
About
Relational graph learning models relational databases as graphs and has demonstrated superior performance on a wide range of relational predictive tasks. However, existing methods struggle to capture long-range dependencies due to information decay in their message-passing mechanisms, and recent relational graph transformers remain limited in jointly modeling structural, semantic, and temporal information. In this paper, we propose GelGT, a Gaussian relational graph transformer that explicitly addresses these challenges. GelGT introduces a structure-semantic collaborative sampling strategy to preserve structural connectivity while filtering irrelevant semantic information, and incorporates a Gaussian graph attention mechanism with a learnable Gaussian bias on the sampled subgraphs to dynamically encode temporal dependencies. Extensive experiments on various real-world datasets demonstrate that GelGT achieves state-of-the-art downstream task performance, with up to a 13.8% improvement in predictive performance.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Driver Top 3 Prediction | rel-f1 | ROC-AUC84 | 70 | |
| Entity Classification | RELBENCH rel-avito user-clicks (test) | AUROC68 | 22 | |
| Entity Classification | RELBENCH rel-stack user-badge (test) | AUROC0.9 | 18 | |
| ad-ctr | RELBENCH rel-avito (test) | MAE0.0362 | 16 | |
| user-attendance | RELBENCH rel-event (test) | MAE0.2423 | 16 | |
| driver-position | RELBENCH rel-f1 (test) | MAE3.7345 | 16 | |
| user-ltv | RELBENCH rel-amazon (test) | MAE14.2538 | 16 | |
| item-sales | RELBENCH rel-hm (test) | MAE0.056 | 16 | |
| study-out binary classification | rel trial (test) | AUROC72 | 12 | |
| item-ltv node regression | rel-amazon RelBench (test) | MAE49.2109 | 7 |