yaochenzhu/PSF-VAE

Official codes of "Path-Specific Counterfactual Fairness for Recommender System (PSF-VAE)". (KDD'23)

PSF-VAE: Path-Specific Counterfactual Fairness for Recommender Systems

The codes are associated with the following paper:

Path-Specific Counterfactual Fairness for Recommender Systems,
Yaochen Zhu, Jing Ma, Liang Wu, Qi Guo, Liangjie Hong, Jundong Li,
SIGKDD 2023.

1. Environments

The codes are written in Python 3.6.5.

• numpy == 1.16.3
• pandas == 0.21.0
• tensorflow-gpu == 1.15.0
• tensorflow-probability == 0.8.0

2. Dataset Acquirement and Simulation

2.1 For the Amazon-VG and ML-1M datasets:

• Acquire the Amazon-VG and ML-1M datasets:
  The original datasets can be found [here] and [here].
  Preprocess the data with data_sim/raw/prepare_data.py.

• Preprocess the original dataset:
  cd to data_sim/raw folder, run
  python prepare_data.py --dataset Name --simulate exposure.

• Fit the binarized rating distributions via VAEs:
  cd to data_sim folder, run
  python train.py --dataset Name --simulate exposure.

• Simulate the datasets:
  python simulate.py --dataset Name.

• The simulated datasets are in psf-vae/data folder

2.2 For the LinkedIn Datasets:

This for now is a private dataset.

3. Model Training and Evaluation

• Split the simulated datasets into train/val/test: cd to psf-vae/data folder, run
  python preprocess.py --dataset Name --split 10.

• Train the PSF-VAE model:
  python train.py --dataset Name --split [0-9]

• Evaluate the model based on recommendation performance and fairness:
  python eval.py --dataset Name --split [0-9]

• Summarize the results into a table:
  python summarize.py --dataset Name

Reference

@inproceedings{Zhu2023Path,  
  title={Path-Specific Counterfactual Fairness for Recommender Systems},      
  author={Zhu, Yaochen and Ma, Jing and Wu, Liang and Guo, Qi and Hong, Liangjie and Li, Jundong},    
  booktitle={Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’23)},    
  year={2023},
}