Graph Neural Network-Based Efficient Subgraph Embedding Method for Link Prediction in Mobile Edge Computing.

Area of Science:

• Network Science
• Graph Theory
• Computer Networks

Background:

• Link prediction is crucial for understanding network evolution and optimizing structures like mobile edge computing (MEC) routing in 5G/6G networks.
• Traditional link prediction methods rely on node similarity, limiting their applicability and generality across diverse network structures.
• Existing algorithms often require predefined functions and struggle with dynamic network changes.

Purpose of the Study:

• To propose efficient and generalizable link prediction algorithms, PLAS and PLGAT, addressing the limitations of traditional methods.
• To introduce a novel approach based on analyzing subgraphs around target node pairs for link prediction.
• To enhance link prediction accuracy and applicability in complex network environments, including 5G/6G MEC networks.

Main Methods:

• Developed PLAS (Predicting Links by Analysis Subgraph) and its Graph Neural Network (GNN) variant, PLGAT (Predicting Links by Graph Attention Networks).
• Implemented an automatic learning of graph structure characteristics by extracting h-hop subgraphs for target node pairs.
• Utilized subgraph information to predict the likelihood of a link existing between the target node pair.

Main Results:

• Experiments on eleven real-world datasets demonstrated the superiority of PLAS and PLGAT over existing link prediction algorithms.
• The proposed methods showed high performance across various network structures, indicating strong generalizability.
• Notably higher Area Under Curve (AUC) values were achieved on 5G MEC Access networks datasets.

Conclusions:

• PLAS and PLGAT offer efficient and versatile solutions for link prediction, outperforming traditional methods.
• The subgraph analysis approach effectively captures network structure characteristics for accurate link prediction.
• These algorithms are particularly promising for applications in evolving networks like 5G/6G MEC, aiding in throughput guidance and node selection.