Overlapping Community Detection in Vehicular Social Networks Based on Graph Attention Autoencoder.

Area of Science:

• Vehicular Social Networks
• Network Science
• Data Mining

Background:

• Community detection is crucial for vehicular social networks, aiding in communication efficiency and privacy.
• Existing methods often overlook overlapping communities and node attribute information, focusing solely on network topology.

Purpose of the Study:

• To propose a novel semi-supervised overlapping community detection method for vehicular social networks.
• To address limitations of existing methods by incorporating both topological and attribute information.

Main Methods:

• Developed a Community Detection method using Graph Attention Autoencoder (CDGAAE).
• Employed a graph attention autoencoder module to fuse topological and attribute data.
• Integrated a modularity optimization enhancement module for overlapping community structures.
• Utilized a semi-supervised clustering module with prior information for enhanced accuracy.

Main Results:

• CDGAAE successfully fuses network topology and node attribute information.
• The method effectively captures overlapping community structures.
• Experimental results demonstrate superior performance of CDGAAE over competing methods on real and synthetic datasets.

Conclusions:

• The proposed CDGAAE method offers an effective approach for semi-supervised overlapping community detection in vehicular social networks.
• This advancement improves communication efficiency, resource allocation, and privacy protection within these networks.