Learning to Map Social Network Users by Unified Manifold Alignment on Hypergraph.

Area of Science:

• Computer Science
• Data Science
• Network Analysis

Background:

• Users maintain multiple social network accounts (e.g., Facebook, Twitter).
• Cross-network user mapping is crucial for applications like personalized recommendations.
• Existing methods lack a unified framework to exploit high-order relational information from social structures and user profiles.

Purpose of the Study:

• To propose a unified hypergraph learning framework (UMAH) for effective cross-network user mapping.
• To automatically determine the weights of profile hyperedges within the unified framework.
• To enhance semisupervised manifold alignment using profile information for accurate user correspondence.

Main Methods:

• Developed a unified hypergraph learning framework (UMAH).
• Modeled social structures and user profile relations within a single hypergraph.
• Learned a common subspace by preserving hypergraph structure and labeled user correspondences.
• Employed semisupervised manifold alignment calibrated with profile information.

Main Results:

• UMAH demonstrated effectiveness in mapping users across different social networks.
• Experimental results on three real-world datasets showed superior performance compared to state-of-the-art methods.
• The framework successfully ranked potential user correspondences based on subspace similarity.

Conclusions:

• UMAH provides a robust and unified approach for cross-network user mapping.
• The framework's ability to integrate social and profile information enhances mapping accuracy.
• This research contributes a novel solution for leveraging relational data in social network analysis.