AttriReBoost: A Gradient-Free Propagation Optimization Method for Cold-Start Mitigation in Attribute Missing Graphs

Area of Science:

• Graph Neural Networks
• Machine Learning
• Data Science

Background:

• Real-world graphs often have incomplete node attributes, degrading the performance of Graph Neural Networks (GNNs).
• Existing feature propagation methods struggle with cold-start problems, especially with attribute resetting and low-degree nodes, hindering convergence.

Purpose of the Study:

• To propose AttriReBoost (ARB), a novel propagation-based method designed to overcome cold-start challenges in attribute-missing graphs.
• To enhance the quality of node representations and improve downstream prediction accuracy in GNNs.

Main Methods:

• AttriReBoost (ARB) redefines initial boundary conditions and incorporates virtual edges to boost global feature propagation (FP).
• The method facilitates gradient-free attribute reconstruction with reduced computational cost.
• Rigorous convergence analysis is provided to ensure method stability.

Main Results:

• ARB achieved an average accuracy improvement of 5.11% over state-of-the-art methods on benchmark datasets.
• The method demonstrated remarkable computational efficiency, processing a 2.44 million-node graph in 16 seconds on a single GPU.

Conclusions:

• AttriReBoost (ARB) effectively mitigates cold-start issues in attribute-missing graphs, significantly improving GNN performance.
• ARB offers a computationally efficient and accurate solution for handling incomplete node attributes in large-scale graph data.