MIGTNet: Metapath Instance-based Graph Transformation Network for heterogeneous graph embedding

Jongmin Park, Soohwan Jeong, Byung Suk Lee, Sungsu Lim

Abstract: Highlights•A new model for heterogeneous graph embedding via metapath instances and relations.•Introducing metapath-instance-based transformation to adopt graph neural networks.•Constructing hierarchical graph attention by aggregating metapath instance embedding.•Learning node embeddings efficiently by sampling metapaths within a maximum lengths.•Achieving higher performance than existing models in graph machine learning tasks.