Amer: A New Attribute-Missing Network Embedding Approach.

Area of Science:

• Computer Science
• Data Science
• Network Analysis

Background:

• Network embedding is crucial for analyzing networks with complete attributes.
• Real-world networks often have missing node attributes due to privacy or resource constraints.
• Existing methods for attribute-missing networks separate attribute completion and embedding, leading to suboptimal performance.

Purpose of the Study:

• To propose a unified model that integrates attribute completion and network embedding.
• To address the limitations of separate processing in existing methods for attribute-missing networks.
• To enhance the accuracy and effectiveness of network analysis on incomplete datasets.

Main Methods:

• Developed a unified model where attribute completion and network representation learning are closely intertwined.
• Utilized mutual information maximization to guide attribute completion using network representation.
• Incorporated attribute-structure relationship constraints via a novel generative adversarial networks (GANs) model.

Main Results:

• The proposed unified model significantly outperforms state-of-the-art methods on attribute-missing network embedding.
• Demonstrated superiority across four network analysis tasks: node classification, node clustering, link prediction, and network visualization.
• Empirical results on real-world datasets validate the effectiveness of the integrated approach.

Conclusions:

• The unified model offers a significant advancement in network embedding for networks with missing attributes.
• Intertwining attribute completion and representation learning leads to superior performance in network analysis.
• This work presents the first unified model for attribute-missing network embedding, setting a new benchmark.