A Node-Collaboration-Informed Graph Convolutional Network for Highly Accurate Representation to Undirected Weighted...

Area of Science:

• Graph Representation Learning
• Network Analysis
• Machine Learning

Background:

• Undirected weighted graphs (UWGs) model complex interactions in big data applications.
• Graph Convolutional Networks (GCNs) represent UWGs but often overlook crucial local collaborative information.
• Information loss in existing GCNs hinders accurate representation learning for UWGs.

Purpose of the Study:

• Propose a Node-Collaboration-Informed Graph Convolutional Network (NGCN) for precise UWG representation.
• Address information loss in GCNs by incorporating local collaborative information.
• Enhance the accuracy of UWG representation learning.

Main Methods:

• Incorporated residual connections and weighted propagation into GCN for global graph characteristics.
• Employed symmetric latent factor analysis (SLFA) to learn local collaborative information from node pairs.
• Developed an adaptive fusion strategy for global and local information to achieve accurate UWG representation.

Main Results:

• The proposed NGCN demonstrates high theoretical representation ability for UWGs.
• Empirical studies on six real-world UWGs show NGCN significantly outperforms leading models in missing link estimation.
• The NGCN model exhibits superior accuracy due to its effective modeling of node collaborations.

Conclusions:

• The NGCN model provides a significant advancement in UWG representation learning.
• Capturing local collaborative information is key to improving GCN performance.
• The NGCN's scalability supports future integration with other GCN extensions.