Faster Graph Embeddings via Coarsening

This repository contains the source code for the experiments in Faster Graph Embeddings via Coarsening.

Summary

We investigate how coarsening algorithms based on the Schur complement affect the predictive performance of random walk-based graph embeddings for different learning tasks. Our multi-label vertex classification experiment builds on the framework for NetMF (Qiu et al., WSDM 2018) and compares various one-vs-all logistic regression models. Our link prediction experiment builds on the setup in node2vec (Grover and Leskovec, KDD 2016) and explores the effect of vertex sparsification on AUC scores for several popular link prediction baselines.

Citation

If you find this code useful in your research, we ask that you cite the following paper:

@inproceedings{fahrbach2020faster,
  title={Faster graph embeddings via coarsening},
  author={Fahrbach, Matthew and Goranci, Gramoz and Peng, Richard and Sachdeva, Sushant and Wang, Chi},
  booktitle={International Conference on Machine Learning},
  pages={2953--2963},
  year={2020},
  organization={PMLR}
}