AAGCN: a graph convolutional neural network with adaptive feature and topology learning.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Graph Neural Networks

Background:

• Deep learning, particularly graph neural networks (GNNs), excels at processing graph-structured data.
• Real-world data frequently exhibit sparsity and missing labels, degrading GNN performance and generalization.

Purpose of the Study:

• To enhance the feature extraction and topological information processing capabilities of GNNs.
• To address challenges of data sparsity and missing labels in graph convolutional neural networks.

Main Methods:

• Proposed an Adaptive Feature and Topology Graph Convolutional Neural Network (AAGCN) model.
• Incorporated an adaptive layer for data preprocessing and feature integration.
• Fused hidden features and topological information with original data features and structure for training.

Main Results:

• The adaptive layer effectively preprocesses data and integrates diverse information.
• Node classification experiments on real datasets demonstrated improved performance.
• The AAGCN model successfully addressed data sparsity and enhanced classification accuracy.

Conclusions:

• The AAGCN model effectively extracts hidden features and topological information from graph data.
• The proposed adaptive layer significantly improves the expressive power and classification performance of GNNs.
• This research offers a robust solution for handling sparse and incomplete graph data.