MCoGCN-motif high-order feature-guided embedding learning framework for social link prediction.

Area of Science:

• Social Network Analysis
• Machine Learning
• Data Mining

Background:

• Traditional social link prediction models often neglect high-order network structures.
• Effective extraction and encoding of these features are crucial for improving prediction accuracy.
• Integrating high-order features into prediction models offers significant theoretical and practical value.

Purpose of the Study:

• To propose a novel embedding learning framework for social link prediction.
• To effectively capture and utilize high-order structural information in social networks.
• To enhance the accuracy and performance of social link prediction models.

Main Methods:

• Utilized motif adjacency matrices to capture complex network patterns.
• Employed a propagation process for node embeddings to carry network structural information.
• Designed a simplified attention mechanism to guide adjacency-based embeddings with motif features.
• Optimized node embeddings using a feed-forward neural network.

Main Results:

• The proposed framework effectively addresses challenges with weakly correlated nodes.
• High-order motif features enhance the similarity and predictive power of node embeddings.
• Experimental evaluations on four social networks demonstrated high accuracy.
• The model showed significant advantages in predicting social links.

Conclusions:

• The novel embedding learning framework successfully integrates high-order motif features for social link prediction.
• The approach enhances node representation by leveraging complex network structures.
• The model offers a promising advancement in the field of social network analysis and link prediction.