.png){kind=icon}
.png){kind=icon}
Return
ReST-Pre: Event prediction by spatial-temporal structural replay on generative implicit event pattern induction
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.130461.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Implicit Event Pattern Induction: A probabilistic graph generation framework is proposed, where variational inference bridges explicit event graphs and temporally enhanced dependency patterns. This overcomes representation bias in surface-level connections while preserving evolutionary dynamics. • Topology-Aware Spatial Propagation: By innovating reverse structural tracing with attention-based aggregation, the method achieves holistic dependency integration beyond local neighborhoods, capturing global interaction patterns in complex event systems. • Decay-Resistant Temporal Modeling: A multi-scale dilated architecture is developed to synergistically capture near-term dynamics and long-term associations, effectively alleviating information decay through structural memory replay mechanisms. • Experimental results show that ReST-Pre significantly outperforms baseline models in Hits@1, Hits@3, Hits@10, and MRR metrics, proving its superiority.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available