GANE: A Generative Adversarial Network Embedding.

Area of Science:

• Computer Science
• Machine Learning
• Graph Theory

Background:

• Network embedding generates low-dimensional feature representations for machine learning.
• Existing methods focus on unsupervised structural preservation or supervised by-product embedding.
• A gap exists in unifying these approaches for enhanced network representation learning.

Purpose of the Study:

• To propose a novel Generative Adversarial Network Embedding (GANE) model.
• To leverage multi-output learning by integrating unsupervised and supervised embedding techniques.
• To improve network representation learning for downstream machine learning tasks.

Main Methods:

• Developed a GANE model using a generative adversarial framework for network embedding.
• Employed a generator to create link edges and a discriminator to differentiate real from generated edges.
• Utilized Wasserstein-1 distance for stable generator training and incorporated pairwise vertex connectivity to preserve network structure.

Main Results:

• The GANE model consistently outperformed state-of-the-art solutions on real-world network datasets.
• Significant improvements were observed in link prediction, clustering, and network alignment tasks.
• The extended GANE effectively preserved original network structural information.

Conclusions:

• GANE offers a robust framework for network embedding by combining generative adversarial learning with structural information preservation.
• The proposed model enhances performance across multiple network analysis tasks.
• GANE represents a significant advancement in network representation learning.