Motif-guided Heterogeneous Graph Deep Generation

Area of Science:

• Computer Science
• Data Science
• Graph Theory

Background:

• Real-world complex systems involve diverse objects and relations, often represented by heterogeneous graphs.
• Heterogeneous graphs capture multi-modal interactions but their generation is challenging.
• Existing methods fail to preserve both local semantics and higher-order structural information.

Purpose of the Study:

• To develop an end-to-end framework for generating novel, high-quality heterogeneous graphs.
• To address limitations in existing methods regarding semantic and structural information preservation.
• To provide robust benchmarks for heterogeneous representation learning tasks.

Main Methods:

• Introduced HGEN, a framework incorporating a heterogeneous walk generator.
• Developed a network motif generator to capture higher-order structural distributions.
• Utilized a heterogeneous graph assembler for adaptive graph construction.

Main Results:

• The proposed method theoretically guarantees preservation of local semantics and global heterogeneous distribution.
• Comprehensive experiments demonstrate the effectiveness and efficiency of HGEN.
• Generated graphs serve as valuable benchmarks for downstream tasks.

Conclusions:

• HGEN offers a powerful and efficient solution for generating realistic heterogeneous graphs.
• The framework successfully preserves crucial graph properties often lost in existing methods.
• This work advances the field of heterogeneous graph generation and representation learning.