MACR

This is an implemention for our SIGKDD 2021 paper based on tensorflow

Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System

by Tianxin Wei, Fuli Feng, Jiawei Chen, Ziwei Wu, Jinfeng Yi and Xiangnan He.

Introduction

MACR is a general popularity debias framework based on causal inference and counterfactual reasoning.

Requirements

tensorflow == 1.14
Numpy
python3
Cython

For LightGCN C++ evaluation, please install Cython and do

cd macr_lightgcn
python setup.py build_ext --inplace

Datasets

We use several recommendation datasets in the following format:

train.txt: Biased training data. Each line is user ID, item ID.
tesst.txt: Unbiased uniform test data. Each line is user ID, item ID.

Run the code

For example:

Normal MF:

python ./macr_mf/train.py --dataset addressa --batch_size 128 --cuda 0 --saveID 1 --log_interval 10 --lr 0.001 --train normalbce --test normal

DynInfoNCE MF:

python ./macr_mf/train.py --dataset addressa --batch_size 128 --cuda 0 --saveID 2 --model dynmf --log_interval 10 --lr 0.001 --train normal --test normal

MACR MF:

python ./macr_mf/train.py --dataset addressa --batch_size 128 --cuda 1 --saveID 1 --log_interval 1 --lr 0.001 --check_c 1 --start -1 --end 1 --step 21 --train rubibceboth --test rubi --alpha 1e-3 --beta 1e-3

Normal LightGCN:

python macr_lightgcn/LightGCN.py --data_path data/ --dataset addressa --verbose 1 --layer_size [64,64] --Ks [20] --loss bce --test normal --epoch 2000 --early_stop 1 --lr 0.001 --batch_size 1024 --gpu_id 1 --log_interval 10

DynInfoNCE LightGCN:

python macr_lightgcn/LightGCN.py --data_path data/ --dataset addressa --verbose 1 --layer_size [64,64] --Ks [20] --loss dyninfo  --test normal --epoch 2000 --early_stop 1 --lr 0.001 --batch_size 128 --gpu_id 0 --log_interval 10

MACR LightGCN:

python macr_lightgcn/LightGCN.py --data_path data/ --dataset addressa --verbose 1 --layer_size [64,64] --Ks [20] --loss bceboth --test rubiboth --start 0 --end 50 --step 31 --epoch 2000 --early_stop 1 --lr 0.001 --batch_size 1024 --gpu_id 1 --log_interval 10 --alpha 1e-3 --beta 1e-3

(The range of counterfactual C's value needs to be set.)

More details and changes will be added soon.

Acknowledgement

Very thanks for Chufeng Shi for his help on code and the LightGCN code repo.

Citation

If you find this paper helpful, please cite our paper.

@inproceedings{wei2021model,
  title={Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System},
  author={Wei, Tianxin and Feng, Fuli and Chen, Jiawei and Wu, Ziwei and Yi, Jinfeng and He, Xiangnan},
  booktitle={Proceedings of the 27th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  year={2021}
}