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

Forget Less, Generalize More: Unifying Temporal and Structural Adaptation for Dynamic Graphs

About

Representation learning on dynamic graphs requires capturing complex dependencies that evolve across both time and structure. Existing approaches typically adopt fixed temporal decay schemes or predetermined structural propagation depths, limiting their ability to generalize across graphs with diverse interaction frequencies and topological characteristics. We propose Dual-Scale Retentive Dynamics (DSRD), a unified framework that maintains a retentive representation state encoding both temporal memory and structural context. DSRD introduces two key components: (i) a retentive state with dual-scale adaptation that jointly models temporal dynamics and structural propagation within a single recurrent formulation, and (ii) adaptive decay kernels with learnable time-sensitivity parameters that automatically balance short-term responsiveness and long-term retention based on the underlying interaction patterns. We provide theoretical analysis establishing the equivalence between event-wise parallel aggregation and efficient recurrent state updates, as well as stability and boundedness guarantees for the learned dynamics. Extensive experiments on 14 real-world benchmarks demonstrate that DSRD consistently achieves state-of-the-art performance on both link prediction and node classification tasks, with strong generalization across transductive and inductive settings.

Qian Chang, Ciprian Doru Giurcaneanu, Runsong Jia, Xia Li, Guoping Hu, Xiufeng Cheng, Jinqing Yang, Mengjia Wu, Yi Zhang• 2026

Related benchmarks

TaskDatasetResultRank
Node ClassificationREDDIT--
216
Inductive dynamic link predictionReddit (inductive)
AUC-ROC (%)99.94
101
Dynamic Link PredictionWikipedia (inductive)
AP99.55
80
Inductive dynamic link predictionWikipedia (inductive)
AUC-ROC0.9952
80
transductive dynamic link predictionWikipedia
AUC ROC99.74
76
transductive dynamic link predictionREDDIT
AUC-ROC0.9998
69
transductive dynamic link predictionCan. Parl.
AUC ROC0.9998
66
Dynamic Link PredictionLastFM (transductive)--
65
transductive dynamic link predictionUCI
AUC96.86
63
transductive dynamic link predictionENRON
AUC93
63
Showing 10 of 50 rows

Other info

Follow for update