Implementation for the PREPRec paper, accepted at Recsys 2024. Our method enables cross-domain, cross-user zero-shot transfer competitive with in-domain SOTA models.

Quick start: follow the instructions in data folder for getting dataset and preprocessing. Then create a res folder to hold trained models and logs of results, and see sample.sh for examples for running and evaluating models.

Coming soon: env dependencies and training hyperparams used in our paper.

Code credits: Original code is based off this pytorch SASRec implementation, with code also taken/repurposed from here, here, here and here.

Please cite our work if you use it:

@misc{wang2024pretrainedsequentialrecommendationframework,
      title={A Pre-trained Sequential Recommendation Framework: Popularity Dynamics for Zero-shot Transfer}, 
      author={Junting Wang and Praneet Rathi and Hari Sundaram},
      year={2024},
      eprint={2401.01497},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2401.01497}, 
}