CAMU: Cycle-Consistent Adversarial Mapping Model for User Alignment Across Social Networks.

Area of Science:

• Computer Science
• Social Network Analysis

Background:

• User alignment across networks is crucial for social network analysis.
• Existing embedding-based methods struggle with data sparsity, noise, and distribution discrepancies, leading to overfitting.
• Limited labeled data hinders the performance of current mapping functions.

Purpose of the Study:

• To propose a cycle-consistent adversarial mapping model for robust user alignment.
• To address representation distribution discrepancies and reduce reliance on labeled data.
• To improve the accuracy and efficiency of cross-network user correspondence.

Main Methods:

• Developed a cycle-consistent adversarial mapping model for latent space alignment.
• Employed adversarial training with discriminators to manage distribution discrepancies.
• Incorporated cycle-consistency training to enhance mapping robustness.
• Utilized both labeled and unlabeled users during the training process.

Main Results:

• The proposed model effectively establishes user correspondence across social networks.
• Demonstrated improved performance compared to existing methods in extensive experiments.
• Showcased the model's ability to handle data sparsity and distribution discrepancies.
• Reduced the requirement for labeled user pairs, mitigating overfitting.

Conclusions:

• The cycle-consistent adversarial mapping model offers a significant advancement in user alignment.
• The approach is effective for real-world social network data.
• This method provides a more robust and data-efficient solution for cross-network user analysis.