Unifying Node Labels, Features, and Distances for Deep Network Completion.

Area of Science:

• Graph theory
• Network science
• Machine learning

Background:

• Real-world network data frequently suffers from incompleteness, including missing nodes and edges.
• Network completion is crucial for accurate analysis and downstream tasks.
• Existing network recovery methods have not fully exploited available information.

Purpose of the Study:

• To propose a novel unified deep graph convolutional network for network completion.
• To effectively infer missing edges by integrating node labels, features, and distances.
• To enhance the exploitation of potential information in network recovery.

Main Methods:

• Constructing an estimated network topology for unobserved parts using node labels.
• Jointly refining network topology and learning edge likelihood.
• Utilizing node labels, node features, and distances in the deep graph convolutional network.

Main Results:

• The proposed method demonstrates superior performance in network completion.
• Extensive experiments on real-world datasets validate the effectiveness of the approach.
• The method outperforms current state-of-the-art network recovery techniques.

Conclusions:

• The unified deep graph convolutional network offers an effective solution for network completion.
• Leveraging node labels, features, and distances significantly improves the inference of missing edges.
• This work advances the field of network recovery by providing a more comprehensive approach.