Detecting influential nodes with topological structure via Graph Neural Network approach in social networks

Area of Science:

• Graph Neural Networks
• Network Science
• Machine Learning

Background:

• Identifying influential nodes is critical in large, dynamic social networks.
• Current methods often focus on either network topology or node features, neglecting a holistic approach.
• A comprehensive evaluation requires considering both structural and attribute information for node relevance.

Purpose of the Study:

• To develop a deep learning framework, DeepInfNode, for identifying influential nodes in graph-based datasets.
• To integrate structural centrality and contextual information for superior node representation.
• To enhance the accuracy and precision of influential node detection in complex networks.

Main Methods:

• Utilized Graph Convolutional Networks (GCN) as the core deep learning architecture.
• Developed the DeepInfNode framework to identify significant nodes by analyzing graph structures.
• Incorporated contextual information from Susceptible-Infected-Recovered (SIR) model simulations to derive node representations and infection rates.

Main Results:

• The DeepInfNode model demonstrated superior performance with higher F1 and Area Under the Curve (AUC) scores.
• Experimental results confirmed the model's effectiveness and precision in identifying key nodes and suggesting new connections.
• Achieved up to 99.1% accuracy improvement compared to state-of-the-art approaches on standard graph datasets.

Conclusions:

• DeepInfNode offers a highly effective and precise method for influential node detection in complex networks.
• The model's ability to integrate network topology and node attributes provides a significant advantage over existing techniques.
• This approach advances the field of network analysis by improving the identification of critical nodes for various applications.