Boosting Temporal Graph Learning From Perspectives of Global and Local Structures.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Graph Neural Networks

Background:

• Temporal graph learning is crucial for applications requiring analysis of dynamic network structures.
• Existing methods often focus on either global or local graph perspectives, limiting their ability to capture complex temporal dynamics.
• A gap exists in effectively integrating complementary global and local information for robust temporal graph representation learning.

Purpose of the Study:

• To propose the Global and Local Embedding Network (GLEN) for effective and efficient temporal graph representation learning.
• To address the limitations of existing methods in capturing complex dynamic patterns by considering both global and local node information.
• To improve the performance of temporal graph learning tasks such as link prediction and dynamic node classification.

Main Methods:

• GLEN dynamically generates node embeddings by integrating global and local perspectives using specialized modules.
• A cross-perspective fusion module combines global and local embeddings to capture high-order semantic relationships.
• The network is designed for efficient computation and effective representation learning on temporal graphs.

Main Results:

• GLEN significantly outperforms baseline methods on multiple real-world datasets for both link prediction and dynamic node classification tasks.
• Experimental results validate the effectiveness of integrating global and local perspectives for temporal graph learning.
• The proposed method demonstrates superior performance under stringent evaluation protocols.

Conclusions:

• GLEN provides a novel and effective approach to temporal graph representation learning by synergizing global and local node embeddings.
• The method achieves a superior balance between inference precision and training efficiency.
• GLEN offers a promising solution for complex dynamic network analysis and various downstream applications.