Node Pair Information Preserving Network Embedding Based on Adversarial Networks.

Area of Science:

• Graph theory
• Machine learning
• Network science

Background:

• Network embedding learns low-dimensional node representations.
• Existing methods often rely on global edge statistics.
• A need exists for methods preserving local node pair information.

Purpose of the Study:

• Propose a novel network embedding model, Node Pair Information Preserving Network Embedding (NINE).
• Utilize local node pair information instead of global statistics.
• Leverage adversarial networks for effective network representation learning.

Main Methods:

• NINE employs an architecture with an NI embedder, NI generator, and NI discriminator.
• The NI embedder uses direct neighbor information and edge existence as labels.
• Generative Adversarial Networks (GANs) ensure the generated NI vectors match embedded distributions.

Main Results:

• Extensive experiments were conducted on seven real-world datasets.
• The model was evaluated on network reconstruction, link prediction, and node classification tasks.
• NINE demonstrated superior effectiveness, efficiency, and rationality compared to seven state-of-the-art models.

Conclusions:

• NINE effectively preserves local node pair information for network embedding.
• The adversarial network approach enhances the quality of learned representations.
• NINE offers a robust and efficient solution for various network analysis tasks.