Graph convolutional networks fusing motif-structure information.

Area of Science:

• Graph Neural Networks
• Deep Learning
• Network Science

Background:

• Big data generates vast graph data, challenging traditional deep learning models.
• Graph neural networks (GNNs) address deep learning limitations with graph data.
• Existing graph convolutional networks (GCNs) primarily use first-order neighbor information, neglecting higher-order structures.

Purpose of the Study:

• To develop a GCN model that captures richer graph topology.
• To mine higher-order information within complex networks.
• To enhance the capability of GCNs by integrating motif-structure information.

Main Methods:

• Proposed a novel GCN model that fuses motif-structure information.
• Identified motif-structures within the network.
• Utilized motif-structure information to refine node aggregation feature weights.

Main Results:

• The proposed GCN model successfully integrates higher-order network information.
• Experiments demonstrated improved node classification accuracy.
• The fusion of motif-structure information enhances GCN performance.

Conclusions:

• The novel GCN model effectively leverages higher-order graph topology.
• Integrating motif-structure information is a viable strategy for improving GCNs.
• The enhanced GCN model shows significant improvements in node classification tasks.