KnowFlow: Empowering decision-making on networks with knowledge-streamlined agent.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Network Science

Background:

• Graph Neural Networks (GNNs) are crucial for decision-making on graph-structured data.
• Real-world GNN applications vary significantly in tasks and graph characteristics.
• Existing GNN methods lack explicit guidelines for incorporating domain knowledge.

Purpose of the Study:

• To address the gap in knowledge integration for GNN design.
• To propose a method that leverages graph learning knowledge to empower GNN development.
• To enhance problem understanding, GNN design, and evaluation using extracted knowledge.

Main Methods:

• Gathering diverse graph learning resources.
• Developing a Large Language Model (LLM)-based agent for knowledge extraction and retrieval.
• Designing four graph learning agents to utilize knowledge across GNN procedures.
• Introducing the KnowFlow framework for knowledge-guided GNN design.

Main Results:

• KnowFlow was evaluated on twelve datasets across node classification, graph classification, and link prediction tasks.
• The method achieved superior performance compared to existing baselines.
• KnowFlow demonstrated effectiveness and efficiency with comparable resource costs.

Conclusions:

• KnowFlow successfully integrates graph learning knowledge into GNN design.
• The proposed approach enhances GNN performance and efficiency across diverse applications.
• This framework offers explicit guidelines for knowledge incorporation in GNN development.