shawn-dm / cafin

This is the official implementation of our ECAI 2023 paper, CAFIN ☕️: Centrality Aware Fairness inducing IN-processing for Unsupervised Representation Learning on Graphs.

CAFIN

CAFIN: Centrality Aware Fairness inducing IN-processing for Unsupervised Representation Learning on Graphs (Arvindh, Aakash, Amul, et. al, ECAI 2023): https://arxiv.org/abs/2304.04391

Overview

The overall implementation is split task-wise,

• src/nc/ contains Node Classification and src/lp/ contains Link Prediction
• approximate_distances.py, centrality_measures.py, dist.py and graph_division.py contain the necessary preprocessing steps.
• experiments.py contains the modified loss functions of CAFIN (Exp 17), CAFIN-N (Exp 18) and CAFIN-P (Exp 19)
• nc/imparity.py contains the implementation of weighted imparity
• utils.py contains necessary supporting functions
• train.py trains CAFIN-GraphSAGE
• train_approx.py uses approximate distances for training
• lr.py evaluates the generated embeddings Overall pipeline can be run using src/nc/run.sh and src/lp/run.sh with appropriate variables set as required.

Reference

If you use CAFIN in your research, please consider citing the following

@misc{
    arun2023cafin,
    title={CAFIN: Centrality Aware Fairness inducing IN-processing for Unsupervised Representation Learning on Graphs}, 
    author={Arvindh Arun and Aakash Aanegola and Amul Agrawal and Ramasuri Narayanam and Ponnurangam Kumaraguru},
    year={2023},
    eprint={2304.04391},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}