Multi-grained alignment method based on stable topics in cross-social networks.

Area of Science:

• Social Network Analysis
• Data Mining
• Computational Social Science

Background:

• Cross-social network user alignment faces challenges with privacy policies hindering full user feature extraction.
• Existing group alignment methods suffer from low accuracy due to a large number of edge users.

Purpose of the Study:

• To develop a novel method for accurate and multi-granularity user alignment across social networks.
• To address the limitations of privacy protection and edge user inaccuracies in current alignment techniques.

Main Methods:

• Utilizing user-generated content (UGC) to extract stable topics via embedded topic jitter time.
• Updating user edge weights using vector distances and employing an improved Louvain algorithm (Stable Topic-Louvain, ST-L) for multi-level community detection.
• Implementing iterative alignment from coarse-grained communities to fine-grained sub-communities for user-level alignment.

Main Results:

• The Stable Topic-Louvain (ST-L) algorithm effectively performs multi-level community detection without predefined tags.
• The iterative alignment process successfully resolves the low accuracy issue associated with edge users at a single granularity.
• The proposed method demonstrates improved accuracy in cross-social network user alignment.

Conclusions:

• The developed ST-L method offers a robust solution for accurate user alignment in cross-social network scenarios.
• The multi-granularity alignment approach enhances precision by iteratively refining community structures.
• The effectiveness of the method is validated through experiments on real-world datasets.