Yifang-Qin/AD-AUG

Implementation of AD-AUG and base models

AD-AUG

This is the pytorch implementation for our ECML-PKDD 2022 paper:

Yifan Wang, Yifang Qin, Yu Han, Mingyang Yin, Jingren Zhou, Hongxia Yang, and Ming Zhang(2022). AD-AUG: Adversarial Data Augmentation for Counterfactual Recommendation

In this paper, we propose a novel counterfactual data augmentation framework, AD-AUG, to mitigate the impact of the imperfect training data and empower CF model.

Please cite our paper if you use the code.

Environment Requirement

The code has been tested running under Python 3.9.7. The required packages are as follows:

• pytorch == 1.7.1
• sklearn == 1.0.2
• pandas == 1.3.5
• numpy == 1.21.2

File Structure

AD-AUG
├── data
└── model
    ├── AD-CDAE
    ├── AD-MacridVAE
    ├── AD-MultVAE
    ├── CDAE
    ├── MacridVAE
    ├── MultVAE
    ├── SLIM
    └── WMF

As listed above, there are two main directories that contain the codes for data preprocessing and model experiments respectively.

Under ./model, we implement all the compared baselines and the AD-AUG augmented version of the AE-based baselines X under the directory named by AD-X,

Running Example

For example, to generate ML-1M data for AD-AUG models, run:

cd ./data && mkdir ml-1m
python process_ml1m.py

which will generate three .csv data under the directory ./data/ml-1m.

To conduct experiment on the Data-oriented MacridVAE, run:

cd ./model/AD-MacridVAE
python main.py --data ../../data/ml-1m --batch 1024 --intern 10 --type data --rg_aug 1e-2

Normally when running Model-oriented methods, AD-AUG requires a larger rg_aug (e.g. 1e4) to limit the output of the augmenter.