Reproduce “A Survey on Heterogeneous Graph Embedding: Methods, Techniques, Applications and Sources” on OpenResearch

About this paper

Xiao Wang, Deyu Bo, Chuan Shi, Shaohua Fan, Yanfang Ye, Philip S. Yu

This survey paper from researchers at Beijing University of Posts and Telecommunications, Case Western Reserve University, and University of Illinois at Chicago provides a comprehensive review of heterogeneous graph embedding, categorizing methods, analyzing their applicability, and detailing resources and future research directions.

Key insights

• Heterogeneous graph embedding methods can be broadly categorized by their primary focus: preserving structural patterns (e.g., meta-paths), integrating attributes (e.g., via HGNNs), tailoring to specific applications, or handling dynamic graph changes.
• A clear distinction exists between shallow HG embedding models (good parallelizability, transductive, high memory) and deep models (inductive, lower parameter memory, end-to-end trainable, often higher computational cost).
• HG embedding techniques have found successful deployment in diverse industrial applications, including e-commerce (recommendation, fraud detection) and cybersecurity (malware detection, key player identification).

Summary by alphaXiv.