Reproduce “Metapath-based Hyperbolic Contrastive Learning for Heterogeneous Graph Embedding” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jongmin Park, Seunghoon Han, Won-Yong Shin, Sungsu Lim

MHCL represents each metapath in its own learnable hyperbolic space and uses contrastive learning to preserve differences among path representations before attention combines them into node embeddings. The method is evaluated on node classification and clustering across IMDB, DBLP, and ACM, and link prediction on LastFM. On ACM clustering, it reaches 76.62 ± 0.81 NMI versus 73.33 ± 0.79 for MSGAT; overall, it outperforms baselines in most cases.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.