Attending Over Triads for Learning Signed Network Embedding.

Area of Science:

• Computer Science
• Network Analysis
• Machine Learning

Background:

• Network embedding learns node representations for downstream tasks.
• Existing methods often focus on single-edge networks, neglecting signed networks.
• Signed networks contain edges with opposite relationships, common in real-world scenarios.

Purpose of the Study:

• To propose a novel network embedding approach for signed networks.
• To address limitations of existing methods that rely solely on local structural information.
• To leverage higher-order network structures for improved node representation learning.

Main Methods:

• Proposed TEA (Triad+Edge+Attention) for signed network embedding.
• Incorporated Structural Balance Theory to utilize higher-order structures (triangles).
• Developed an attention mechanism to weigh the importance of different triangles.

Main Results:

• TEA effectively learns node representations in signed networks.
• The method demonstrated superior performance compared to strong baseline approaches.
• Experiments on real-world signed networks validated TEA's effectiveness.

Conclusions:

• TEA offers a significant advancement in signed network embedding.
• Leveraging higher-order structures and attention enhances prediction accuracy.
• The proposed method is effective for analyzing complex signed network data.