BiVAE: Bilateral Variational Autoencoder

This is the code for the paper:

Bilateral Variational Autoencoder for Collaborative Filtering


Quoc-Tuan Truong, Aghiles Salah, and Hady W. Lauw


Presented at WSDM 2021

If you find the code and data useful in your research, please cite:

@inproceedings{truong2021bilateral,
  title={Bilateral Variational Autoencoder for Collaborative Filtering},
  author={Truong, Quoc-Tuan, and Salah, Aghiles, and Lauw, Hady W},
  booktitle={ACM International Conference on Web Search and Data Mining, {WSDM} 2021}
  year={2021},
}

How to run

pip install -r requirements.txt

Run BiVAE model:

python bivae.py -d office -k 20 -e '[40]' -a tanh -l pois -ne 500 -bs 128 -lr 0.001 -tk 50 -v

Run BiVAE model with Constrained Adaptive Priors (CAP):

• CAP requires feature learning, here we use vanilla VAE as an example:
  ```bash

  user side

  python feature_learning.py -d office -w user -k 20 -e ‘[100]’ -a tanh -l pois -ne 100 -bs 128 -lr 0.001 -s 123 -v

item side

python feature_learning.py -d office -w item -k 20 -e ‘[100]’ -a tanh -l pois -ne 100 -bs 128 -lr 0.001 -s 123 -v

- When user/item features are ready, we can train BiVAE with CAP:
```bash
python bivae_cap.py -d office -uc -ic -k 20 -e '[40]' -a tanh -l pois -ne 500 -bs 128 -lr 0.001 -tk 50 -v

Related

• Implementation of BiVAE in Cornac
• BiVAE tutorial with Microsoft Recommenders

Contact

Questions and discussion are welcome: www.qttruong.com