DyGCN: Efficient Dynamic Graph Embedding With Graph Convolutional Network.

Area of Science:

• Computer Science
• Machine Learning
• Graph Neural Networks

Background:

• Graph embedding learns low-dimensional node representations.
• Graph convolutional networks (GCNs) are effective but primarily for static graphs.
• Dynamic graphs require efficient embedding methods that capture evolving structures.

Purpose of the Study:

• To propose an efficient dynamic graph embedding method, dynamic GCN (DyGCN).
• To extend GCN-based methods for handling dynamic graph structures.
• To enable efficient and performance-preserving updates of node embeddings in evolving graphs.

Main Methods:

• Generalized GCN embedding propagation to a dynamic setting.
• Developed an efficient propagation scheme for topological and neighborhood embedding changes.
• Implemented an update strategy prioritizing the most affected nodes.

Main Results:

• DyGCN efficiently updates node embeddings in dynamic graphs.
• The model preserves performance while significantly saving time.
• Experimental validation on various dynamic graphs confirmed effectiveness.

Conclusions:

• DyGCN offers an efficient solution for dynamic graph embedding.
• The proposed method effectively handles evolving graph structures.
• DyGCN balances computational efficiency with embedding accuracy.