.png){kind=icon}
Return
DAGE: decay-aware graph extrapolation for temporal knowledge graph reasoning
.png){kind=icon}
PRE
AI
DOI：10.1016/j.inffus.2026.104484.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Temporal knowledge graph extrapolation aims to reason about and predict future unknown facts. Current mainstream approaches are predominantly path-based, which model event evolution patterns through multi-hop temporal paths between entities. Most approaches neglect the temporal decay effect between historical events and query events, which introduces noise and hinders the model's ability to focus on recent evidence of causal significance. This results in inadequate modelling capabilities for long-term sequential dependencies. This issue fundamentally stems from an inadequate integration of information. To address this, we propose a novel Decay-Aware Graph Extrapolation method (DAGE). This approach introduces temporal decay as a guiding principle for fusion. We first devise a hierarchical temporal window structure to reconstruct the temporal graph, reducing redundant paths while capturing temporal decay effects. Secondly, we introduce a cross-modal fusion module that dynamically intertwines relational embeddings with time distance features to distinguish time distance variations between events. Finally, we incorporate a temporal decay-enhanced information aggregation module to balance the influence of events across different temporal paths on the query event. Experimental results demonstrate that the model achieves competitive performance across four commonly used public benchmark datasets. Furthermore, the model exhibits remarkable generalization capacity in cross-time forecasting scenarios. These findings collectively validate the significant enhancement of reasoning performance through temporal decay modelling. Our code is available at https://github.com/MZ-MiaoZhang/DAGE.
Keywords:
Temporal knowledge graph
Graph extrapolation
Temporal decay
Knowledge graph reasoning
Journal
IF:
15.5
Papers:
4.2K
Citations:
2.7W
Organization
Cited Papers
No cited papers available