Parameter-Efficient Variational AutoEncoder for Multimodal Multi-Interest Recommendation

This repository is the official implementation of paper

Nhu-Thuat Tran and Hady W. Lauw. 2025. Parameter-Efficient Variational AutoEncoder for Multimodal Multi-Interest Recommendation. Accepted by The 33rd ACM International Conference on Multimedia (ACMMM'2025), Dublin, Ireland, October 27-31, 2025.

Environment

• Anaconda: 4.12.0
• Python: 3.7.5
• OS: MacOS

Data

Please follow the instruction in README.md file under data folder

Requirements

Create virtual environment

conda create --prefix ./maplevae python=3.7.5 -y

Activate environment

conda activate ./maplevae

Install requirements

pip install -r requirements.txt

Training and Evaluation

1. Create a YAML config file under configs folder as samples.

2. Prepare run.sh file as follows

python run_maplevae.py --dataset <dataset_name> --config_file <your_config_file> --device_id <ID of GPU machine>

3. To run training and evaluation

bash run.sh

Hyper-parameter tuning

We follow instructions from VALID (https://github.com/PreferredAI/VALID) for hyper-parameter tuning.

Then, we tune the key hyper-parameters in MapleVAE

• n_gcn_layers: the number of GCN layers 1, 2, 3, 4
• alpha: multimodality graph combination weight (Section 3.2.3) 0.6, 0.7, 0.8, 0.9
• Q: the number of quantizing levels and W: the size of each codebook in RQ-VAE by changing text_code_file and image_code_file.

Citation

If you find our work useful for your research, please cite our paper as

@inproceedings{MapleVAE,
  author       = {Nhu{-}Thuat Tran and
                  Hady W. Lauw},
  title        = {Parameter-Efficient Variational AutoEncoder for Multimodal Multi-Interest Recommendation},
  booktitle    = {Proceedings of The 33rd ACM International Conference on Multimedia, Dublin, Ireland, October 27-31, 2025},
  year         = {2025}
}