PromptGCN: Bridging Subgraph Gaps in Lightweight GCNs

Area of Science:

• Machine Learning
• Graph Neural Networks
• Deep Learning

Background:

• Graph Convolutional Networks (GCNs) face memory constraints on large graphs, leading to out-of-memory errors.
• Subgraph sampling methods reduce memory usage but create information gaps, decreasing GCN accuracy.
• Existing lightweight GCNs struggle to leverage global graph context effectively.

Purpose of the Study:

• To propose PromptGCN, a novel prompt-based lightweight GCN model.
• To bridge the information gaps inherent in subgraph sampling methods.
• To enhance the accuracy and efficiency of GCNs on large-scale graphs.

Main Methods:

• Introduced learnable prompt embeddings to capture global graph information.
• Attached prompts to each subgraph to facilitate global information transfer.
• Integrated prompt-based learning with subgraph sampling techniques.

Main Results:

• PromptGCN demonstrated superior performance across seven large-scale graph datasets.
• Achieved up to a 5.48% accuracy improvement on the Flickr dataset compared to baseline methods.
• Effectively reduced memory consumption while maintaining or improving accuracy.

Conclusions:

• PromptGCN successfully bridges subgraph information gaps, enhancing GCN accuracy.
• The model is easily integrable with existing subgraph sampling methods.
• PromptGCN offers a viable solution for efficient and accurate large-scale graph analysis.