Encoding edge type information in graphlets.

Area of Science:

• Graph theory and network analysis
• Machine learning and data mining

Background:

• Graph embedding methods are widely used but often overlook crucial edge attribute information.
• Existing methods lack explainability and struggle with edge-attributed networks.

Purpose of the Study:

• To develop a novel framework for embedding edge type information into graph representations.
• To enhance the explainability and utility of graph embeddings for edge-attributed networks.
• To introduce the Typed-Edge Graphlets Degree Vector (TyE-GDV) and extend combinatorial approaches.

Main Methods:

• Developed a framework to embed edge type information in graphlets, creating the TyE-GDV.
• Extended colored graphlets and heterogeneous graphlets approaches for edge-attributed networks.
• Applied methods to a chronic pain patient network case study and node classification tasks.

Main Results:

• Network structure and specific social ties (friends, colleagues, healthcare professionals) are critical indicators of chronic pain severity.
• Edge-type encoded graphlets approaches significantly outperform traditional graphlet degree vectors in node classification.
• TyE-GDV achieves competitive performance with combinatorial approaches while being more space-efficient.

Conclusions:

• Incorporating edge type information enhances graph embedding's effectiveness and explainability.
• The proposed TyE-GDV method provides an efficient and powerful tool for analyzing complex, edge-attributed networks.
• Findings highlight the importance of social network structures in understanding chronic pain management.