EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

AISE: Dynamic Graph Representation Learning With Adaptive Intensity Estimation and Semantic-Aware Sampling

Yanghong Qin

IEEE Access · 2026 · 14 · 94629-94636

DOI: 10.1109/access.2026.3704564 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Yanghong Qin. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Science and Technology Research Program of Chongqing Municipal Education Commission, Chongqing Three Gorges Polytechnic College through the University-Level Research Project.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Representation learning on continuous-time dynamic graphs (CTDGs) is a fundamental research problem in graph learning, whose core challenge lies in the persistent evolution of dynamic interactions. Preserving spatial-temporal proximity among nodes is a critical requirement for CTDG representation learning. However, existing methods suffer from two major limitations: first, interaction intensity between node pairs is typically modeled under rigid parametric assumptions (e.g., Hawkes processes), which struggle to accommodate the diverse interaction patterns across different node pairs; second, the neighbor sampling strategies employed by existing works are decoupled from the semantic information of interaction events, leading to suboptimal context node selection for proximity estimation. To address these issues, we propose AISE, a framework integrating a Pattern-Adaptive Intensity Estimation (PAIE) module and a Semantic-Aware Neighbor Sampling (SANS) strategy. PAIE dynamically infers personalized decay coefficients from historical interaction interval sequences under a non-homogeneous Poisson process assumption. SANS jointly leverages event features and temporal proximity scores to select more informative context nodes via importance-weighted sampling. Experiments on multiple public benchmarks show that AISE outperforms existing baselines on both link prediction and node classification tasks.

Förderung

  • Science and Technology Research Program of Chongqing Municipal Education Commission
  • Chongqing Three Gorges Polytechnic College through the University-Level Research Project

Zitieren

APA

Qin, Y. (2026). AISE: Dynamic Graph Representation Learning With Adaptive Intensity Estimation and Semantic-Aware Sampling. IEEE Access, 14, 94629-94636. https://doi.org/10.1109/access.2026.3704564

Vancouver

Qin Y. AISE: Dynamic Graph Representation Learning With Adaptive Intensity Estimation and Semantic-Aware Sampling. IEEE Access. 2026;14:94629-94636. doi:10.1109/access.2026.3704564

BibTeX
@article{Qin2026,
  author = {Qin, Yanghong},
  title = {AISE: Dynamic Graph Representation Learning With Adaptive Intensity Estimation and Semantic-Aware Sampling},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {94629--94636},
  doi = {10.1109/access.2026.3704564}
}
Kostenlos starten →