Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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.

Zezhong Ding, Jin Li, Xugang Wang, Xike Xie• 2026

Related benchmarks

TaskDatasetResultRank
Driver Top 3 Predictionrel-f1
ROC-AUC84
70
Entity ClassificationRELBENCH rel-avito user-clicks (test)
AUROC68
22
Entity ClassificationRELBENCH rel-stack user-badge (test)
AUROC0.9
18
ad-ctrRELBENCH rel-avito (test)
MAE0.0362
16
user-attendanceRELBENCH rel-event (test)
MAE0.2423
16
driver-positionRELBENCH rel-f1 (test)
MAE3.7345
16
user-ltvRELBENCH rel-amazon (test)
MAE14.2538
16
item-salesRELBENCH rel-hm (test)
MAE0.056
16
study-out binary classificationrel trial (test)
AUROC72
12
item-ltv node regressionrel-amazon RelBench (test)
MAE49.2109
7
Showing 10 of 30 rows

Other info

Follow for update