FairMI

This repository contains source code to our AAAI-2023 paper:

• Fair Representation Learning for Recommendation: A Mutual Information-Based Perspective

FairMI is a framework based Mutual Information (MI) for embedding fairness in recommendations, which is optimized by the two-fold MI based objective.

Note: Due to the limitations of github's file transfer, the complete training parameters are published in https://drive.google.com/drive/folders/14A4SqcPVFxBpFdX7TAL1BP_oShwl3kPK?usp=sharing

Getting Started

Train & Test

• Training BPRMF baseline on MovieLens:

python train_bpr_baseline.py

• Training FairMI_BPR on MovieLens:

python train_bpr_fairmi.py

• Training GCN baseline on MovieLens:

python train_gcn_baseline.py

• Training FairMI_GCN on MovieLens:

python train_gcn_fairmi.py