JORA: Weakly Supervised User Identity Linkage via Jointly Learning to Represent and Align.

Area of Science:

• Computer Science
• Network Science
• Data Mining

Background:

• User identity linkage across networks is crucial for social network applications.
• Existing methods often use separate stages for network embedding and space alignment, leading to conflicting objectives.
• Defining similarities for unlabeled cross-network users and reliance on extensive labeled data are significant challenges.

Purpose of the Study:

• To address limitations in current user identity linkage methods.
• To propose a weakly-supervised approach for cross-network user identity linkage.
• To develop a model that reduces reliance on labeled data and handles unlabeled user pair similarities effectively.

Main Methods:

• Introduced the JOintly learning to Represent and Align (JORA) model.
• Utilized an inductive graph convolutional network (GCN) for network representation learning.
• Implemented joint optimization of representation and alignment learning components.
• Incorporated an attention mechanism for self-adaptive similarity learning in unlabeled user pairs.

Main Results:

• The JORA model demonstrated superior performance in user identity linkage tasks.
• Joint optimization effectively integrated network characteristics into the alignment process.
• The attention mechanism mitigated issues arising from predefined similarities for unlabeled pairs.
• Reduced the requirement for labeled aligned user pairs compared to previous methods.

Conclusions:

• The proposed JORA model offers a robust solution for weakly-supervised user identity linkage.
• Joint learning framework effectively addresses the conflict between representation and alignment objectives.
• The model shows significant improvements over state-of-the-art methods on real-world social networks.