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

2026-05-28

AI Summary

Dynamic graph representation learning struggles to simultaneously adapt to varying interaction frequencies and evolving topologies, with existing methods suffering from limited generalization due to fixed temporal decay schemes or predefined propagation depths. This work proposes the Dual-Scale Memory Dynamics (DSRD) framework, which unifies temporal memory and structural context into a single memory state for the first time. DSRD employs learnable time-sensitive parameters to adaptively balance short-term responsiveness and long-term memory, and introduces an adaptive decay kernel alongside an event-driven parallel aggregation algorithm. Theoretical analysis establishes the equivalence between recurrent updates and event-level aggregation, as well as system stability. Evaluated on 14 real-world dynamic graph benchmarks, DSRD achieves state-of-the-art performance in both link prediction and node classification, demonstrating strong transductive and inductive generalization capabilities.

Abstract

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.

Problem

• dynamic graphs
• representation learning
• temporal adaptation
• structural adaptation
• generalization

Innovation

• dynamic graphs
• temporal adaptation
• structural propagation
• adaptive decay kernels
• retentive representation