A Weighted Symmetric Graph Embedding Approach for Link Prediction in Undirected Graphs

Area of Science:

• Graph embedding
• Network analysis
• Machine learning

Background:

• Link prediction is crucial in social network analysis, with deep learning embeddings showing promise.
• Existing methods face challenges in node embedding (neighbor weighting) and edge embedding (symmetry).

Purpose of the Study:

• To propose a weighted symmetric graph embedding approach to address limitations in current link prediction methods.
• To enhance node and edge embedding techniques for improved link prediction accuracy.

Main Methods:

• Developed a weighted approach for aggregating neighbors in node embedding.
• Implemented bidirectional concatenation for symmetric edge embedding, preserving local structure.

Main Results:

• The proposed method achieved superior performance in link prediction tasks.
• Demonstrated improved accuracy and symmetry in learned edge representations compared to state-of-the-art methods.

Conclusions:

• The weighted symmetric graph embedding approach effectively enhances link prediction.
• Accurate and symmetric edge representations are key to improving network link prediction performance.