Dual-channel deep graph convolutional neural networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Graph Neural Networks

Background:

• Dual-channel graph convolutional neural networks (GCNs) integrate hybrid features for improved machine learning tasks.
• Current dual-channel GCNs face performance limitations due to a restricted number of convolution layers, leading to over-smoothing and decreased efficacy.
• Over-smoothing in deep GCNs diminishes performance as the number of layers increases.

Purpose of the Study:

• To address the over-smoothing issue in dual-channel GCNs.
• To enhance the performance of dual-channel GCNs by increasing model depth.
• To propose a novel dual-channel deep graph convolutional neural network (D2GCN) architecture.

Main Methods:

• Incorporating residual connections, inspired by convolutional neural networks, into dual-channel GCNs.
• Developing a dual-channel deep graph convolutional neural network (D2GCN) model.
• Evaluating D2GCN performance on benchmark datasets for node classification tasks.

Main Results:

• The proposed D2GCN effectively mitigates the over-smoothing phenomenon.
• D2GCN demonstrates superior performance compared to existing algorithms in node classification.
• Experimental validation on CiteSeer, DBLP, and SDBLP datasets confirms D2GCN's effectiveness.

Conclusions:

• Residual connections enable deeper dual-channel GCNs without performance degradation.
• D2GCN offers an effective solution for node classification tasks by overcoming limitations of traditional GCNs.
• The D2GCN architecture represents a significant advancement in graph convolutional neural network research.