This repository hosts the code and experiments for an ongoing research project on heterogeneous graph coarsening. The goal is to develop and evaluate data-driven methods that reduce the size of large-scale heterogeneous graphs while preserving their structural and semantic properties.
This project is still under development. Some components may be incomplete or subject to change. Final results, trained models, and detailed documentation will be provided once the research is finalized.
Traditional graph coarsening methods are limited when applied to heterogeneous graphs—graphs with multiple types of nodes and/or edges. This project aims to:
- Propose novel coarsening techniques for heterogeneous structures.
- Leverage machine learning models to learn coarsening policies.
- Evaluate coarsening impact on downstream tasks (e.g., node classification).