Preserving high-order ego-centric topological patterns in node representation in heterogeneous graph

Tianqianjin Lin, Yangyang Kang, Zhuoren Jiang, Kaisong Song, Xurui Li, Hongsong Li, Jiawei Liu, Changlong Sun, Cui Huang, Xiaozhong Liu

Abstract: Highlights•Learns high-order ego-centric patterns tailored to the graph and task automatically.•Represents nodes by similarity to learned patterns, preserving ego-centric semantics.•Provides built-in interpretability through the learned patterns.•Evaluated on two million-scale and four benchmark datasets, achieving strong results.