[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

When Speed meets Accuracy: an Efficient and Effective Graph Model for Temporal Link Prediction

Created by
  • Haebom

Author

Haoyang Li, Yuming Xu, Yiming Li, Hanmo Liu, Darian Li, Chen Jason Zhang, Lei Chen, Qing Li

Outline

In this paper, we propose EAGLE, a lightweight framework for temporal link prediction in dynamic graphs. While existing Temporal Graph Neural Networks (T-GNNs) suffer from high computational cost due to their complex structures, EAGLE solves this problem by integrating short-term recent neighbor information and long-term global structural patterns. The temporal-aware module aggregates recent neighbor information of nodes, and the structural-aware module captures the influence of globally important nodes by utilizing temporal Personalized PageRank. It uses an adaptive weighting mechanism that dynamically adjusts the contributions of the two modules according to data characteristics, and significantly improves efficiency by eliminating complex multi-stage message passing or memory-intensive mechanisms. Experimental results show that EAGLE outperforms existing state-of-the-art T-GNNs in terms of effectiveness and efficiency, and is more than 50 times faster than transformer-based T-GNNs.

Takeaways, Limitations

•
Takeaways:
◦
Presentation of a lightweight framework that effectively solves the computational cost problem of existing T-GNN.
◦
Improve forecast accuracy by effectively integrating short-term and long-term patterns.
◦
Achieve optimal performance suited to data characteristics through an adaptive weighting mechanism.
◦
More than 50x speed improvement compared to existing state-of-the-art T-GNN.
•
Limitations:
◦
The performance of the proposed EAGLE may be biased towards certain types of dynamic graphs.
◦
Need to verify generalization performance for graphs of various sizes and characteristics.
◦
Further analysis is needed on the complexity of temporal Personalized PageRank calculation.
◦
Further research is needed on its applicability to extremely large graphs.
👍
Made with Slashpage