Anchor Link Prediction across Attributed Networks via Network Embedding.

Area of Science:

• Social Network Analysis
• Machine Learning
• Data Mining

Background:

• Users increasingly participate in multiple social networks.
• Identifying users across these networks (anchor link prediction) is crucial for applications like cross-network recommendations and user profiling.
• Existing methods often rely on hand-crafted features and overlook network attribute information.

Purpose of the Study:

• To propose a novel semi-supervised network-embedding model for accurate anchor link prediction.
• To leverage both network topology and attribute information within a unified framework.
• To improve upon existing techniques by incorporating semi-supervised learning.

Main Methods:

• Developed a semi-supervised network-embedding model.
• Represented each node in multiple networks as a vector.
• Learned embeddings using observed anchor links as supervision, alongside network topology and attributes.

Main Results:

• The proposed model demonstrated superior performance compared to state-of-the-art techniques.
• Experiments were conducted on real-world datasets, validating the model's effectiveness.
• The method successfully integrates structural and attribute information for improved prediction.

Conclusions:

• The novel semi-supervised network-embedding model offers a robust solution for anchor link prediction.
• The approach effectively utilizes multi-network data, including topology and attributes.
• This work advances the field of cross-network user identification.