Multi-Task Learning Based Network Embedding.

Area of Science:

• Computer Science
• Data Science
• Network Analysis

Background:

• Network representation learning aims to embed network structures into low-dimensional spaces.
• Existing methods often use single-task learning, limiting their ability to capture comprehensive node relationships.
• Node proximity depends on both local and global network structures.

Purpose of the Study:

• To propose a novel Multi-Task Learning-Based Network Embedding (MLNE) method.
• To address the limitations of single-task learning in network representation.
• To obtain node embeddings that accurately reflect node roles within networks.

Main Methods:

• Developed MLNE, a multi-task learning framework for network embedding.
• Incorporated two tasks: preserving high-order proximity across the network and low-order proximity within local neighborhoods.
• Utilized a supervised deep learning model for joint task learning.

Main Results:

• MLNE generates node embeddings that better capture node roles.
• Experimental results on five real-world networks demonstrate competitive performance.
• Evaluated on multi-label classification, link prediction, and visualization tasks.

Conclusions:

• MLNE offers a superior approach to network representation learning compared to existing methods.
• The multi-task learning strategy effectively preserves both global and local network information.
• The learned embeddings provide a richer understanding of network topology and node functions.