VARIUM: Variational Autoencoder for Multi-Interest Representation with Inter-User Memory

This repository is the official implementation of paper (https://dl.acm.org/doi/pdf/10.1145/3701551.3703558)

Nhu-Thuat Tran and Hady W. Lauw. 2025. VARIUM: Variational Autoencoder for Multi-Interest Representation with Inter-User Memory. In Proceedings of 18th ACM International Conference on Web Search and Data Mining (WSDM’25), Hannover, Germany, March 10-14, 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 ./varium python=3.7.5 -y

Activate environment

conda activate ./varium

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_varium.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

The base model follows the implementation of VALID (https://github.com/PreferredAI/VALID). Therefore, we first follow the hyper-parameter tuning from VALID, without using memory network by setting use_memory to False.

Then we tune the key hyper-parameters in VARIUM's architecture

• num_steps: the number of refinement layers in 1, 2, 3, 4
• n_memory_blocks: the number of slots in memory 16, 32, 48, 64 (extending this list for your custom datasets might lead to better performance)
• tau_memory: temperature in memory 0.2, 0.3, 0.4, 0.5
• rho_carry: hyper-parameter (Equation 7 in the paper) in range [0.5, 3] with step size 0.5

Citation

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

@inproceedings{VARIUM,
  author       = {Nhu{-}Thuat Tran and
                  Hady W. Lauw},
  title        = {{VARIUM:} Variational Autoencoder for Multi-Interest Representation
                  with Inter-User Memory},
  booktitle    = {Proceedings of the Eighteenth {ACM} International Conference on Web
                  Search and Data Mining, {WSDM} 2025, Hannover, Germany, March 10-14,
                  2025},
  pages        = {156--164},
  year         = {2025}
}