Simple yet effective heuristic community detection with graph convolution network.

Area of Science:

• Graph Neural Networks
• Network Science
• Machine Learning

Background:

• Existing graph neural network (GNN) community detection methods often require predefining the number of communities, introducing bias and complexity.
• Current algorithms frequently rely on contrastive objectives or data augmentation, increasing hyperparameters and computational overhead.
• These limitations hinder the efficiency and generalizability of deep learning approaches for community detection.

Purpose of the Study:

• To develop an adaptive community detection framework for GNNs that eliminates the need for pre-specified community numbers.
• To simplify the training process by removing contrastive learning and data augmentation dependencies.
• To improve the quality, efficiency, and accuracy of community detection in graph data.

Main Methods:

• Introduced an adaptive detection method to identify high-quality structural communities as global references.
• Proposed a novel mechanism for modeling node-community relationships, integrating global, local, and attribute information.
• Applied a reconstructed soft modularity loss for end-to-end optimization of node-community relationships.

Main Results:

• The proposed framework demonstrates superior detection efficiency and competitive accuracy across multiple graph datasets.
• The approach significantly reduces training complexity and prior dependency compared to traditional and recent deep learning methods.
• The method effectively enhances community structure without relying on data augmentation or contrastive learning.

Conclusions:

• The adaptive GNN framework offers an efficient and lightweight solution for high-quality community detection.
• Eliminating pre-specified community numbers and contrastive learning simplifies the process while maintaining performance.
• This approach provides a promising direction for advancing unsupervised community detection in complex networks.