Behavior-Dependent Linear Recurrent Units for Efficient Sequential Recommendation

Behavior-Dependent Linear Recurrent Units for Efficient Sequential Recommendation (CIKM 2024)
Chengkai Liu, Jianghao Lin, Hanzhou Liu, Jianling Wang, James Caverlee
Paper: https://arxiv.org/abs/2406.12580

Usage

Requirements

• Python >= 3.7
• PyTorch >= 1.12
• CUDA >= 11.6
• Triton >= 2.2
• Install RecBole:
  • pip install recbole
• [optional] Install causal Conv1d with CUDA optimization for faster computation of Conv1D:
  • pip install causal-conv1d>=1.2.0

Run

python run.py

Please update config.yaml to adjust the hyperparameters and experimental settings.

Citation

@inproceedings{liu2024behavior,
  title={Behavior-Dependent Linear Recurrent Units for Efficient Sequential Recommendation},
  author={Liu, Chengkai and Lin, Jianghao and Liu, Hanzhou and Wang, Jianling and Caverlee, James},
  booktitle={Proceedings of the 33rd ACM International Conference on Information and Knowledge Management},
  pages={1430--1440},
  year={2024}
}

Acknowledgment

This project references RecBole, Accelerated Scan and Causal-Conv1d. We appreciate their outstanding work and commitment to open source.