Context-Aware Sequential Model for Multi-Behaviour Recommendation

This repository contains the PyTorch implementation of the paper: "Context-Aware Sequential Model for Multi-Behaviour Recommendation" by Shereen Elsayed, Ahmed Rashed, and Lars Schmidt-Thieme. https://arxiv.org/abs/2312.09684

Summary

CASM is a sequential recommendation model that utilizes multiple user behaviors to capture complex patterns in historical interactions. It outperforms state-of-the-art approaches on four real-world datasets by leveraging context-aware multi-head self-attention layers and a weighted binary cross-entropy loss to support an arbitrary number of behaviors.

Performance Boost

Our CASM model shows a performance improvement of up to 19.24% over CARCA on various datasets.

Method	MovieLens HR@10	MovieLens NDCG@10	Tianchi HR@10	Tianchi NDCG@10
Sequential Recommendation Methods
SASRec	0.911 ± 1𝐸−3	0.668 ± 5.1𝐸−3	0.659 ± 3𝐸−3	0.495 ± 2𝐸−3
SSE-PT	0.911 ± 7.1𝐸−3	0.657 ± 4.5𝐸−3	0.663 ± 1.2𝐸−2	0.468 ± 1.3𝐸−2
Context-Aware Recommendation Methods
CARCA	0.906 ± 2𝐸−3	0.665 ± 1𝐸−3	0.713 ± 4𝐸−4	0.500 ± 1𝐸−3
Multi-Behavior Recommendation Methods
MATN	0.847	0.569	0.714 ± 7𝐸−4	0.485 ± 2𝐸−3
KHGT	0.861	0.597	0.652 ± 1𝐸−4	0.443 ± 1𝐸−4
MBHT	0.913 ± 5.9𝐸−3	0.695 ± 7𝐸−3	0.725 ± 6.3𝐸−3	0.554 ± 4.8𝐸−3
CASM	0.930 ± 6𝐸−4	0.713 ± 1.3𝐸−3	0.755 ± 9𝐸−4	0.584 ± 2.7𝐸−3
Improv.(%)	1.86%	2.44%	2.99%	5.95%

Usage

1. Clone the repository:

   git clone https://github.com/ariaattar/CASM-PyTorch.git
   cd CASM-PyTorch

2. Clone the repository:

   unzip data/Tmall.zip -d data

3. Install the required packages:

   pip install -r requirements.txt

4. Train the model:

   python main.py

Original TensorFlow Implementation

For the original TensorFlow implementation of the CASM model, please visit the following repository: CASM TensorFlow Implementation

Citation

If you use this code for your research, please cite our paper:

@article{elsayed2023casm,
  title={Context-Aware Sequential Model for Multi-Behaviour Recommendation},
  author={Elsayed, Shereen and Rashed, Ahmed and Schmidt-Thieme, Lars},
  journal={arXiv preprint arXiv:2312.09684},
  year={2023}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.