DCBR (AAAI'25)

PyTorch implementation for DCBR proposed in the following paper:

Disentangled Contrastive Bundle Recommendation with Conditional Diffusion
Jiuqiang Li
In AAAI 2025
Paper

Overview

Environment

We implement our code in the following environment.

• OS: Ubuntu 20.04
• GPU: NVIDIA RTX 3090(24GB) * 1
• CPU: 14 vCPU Intel(R) Xeon(R) Platinum 8362 CPU @ 2.80GHz
• Python 3.8.10
• torch==1.11.0+cu113
• numpy==1.22.4
• PyYAML==6.0.2
• scipy==1.10.1
• tqdm==4.61.2

Run the following command to install the dependencies.

pip install -r requirements.txt

Dataset

The datasets utilized in the experiments include $MealRec_H^+$, $MealRec_L^+$, and iFashion. $MealRec^+$ and iFashion have been published in the MealRecPlus and CrossCBR codebases, respectively.

Training

The command to train DCBR on the $MealRec_H^+$ / $MealRec_L^+$ / iFashion dataset is as follows.

python train.py -m DCBR -d {dataset_name}

Please set the hyperparameters in the configs/models/DCBR.yaml file.

Evaluation

The command to evaluate the performance of the pretrained checkpoints on the specified dataset is as follows.

python test.py -m DCBR -d {dataset_name} -c {checkpoint_path}

In test.py, two evaluation methods are provided. Specifically:

• quick_test(): Directly calculates metrics using the pre-trained embeddings from the checkpoints.
• test(): Infers to obtain embeddings before calculating metrics.

Reproducibility

We report the best configuration, training logs, and corresponding checkpoints of DCBR to reproduce the results in Table 2 of our paper.

Dataset	Download
MealRecH+	conf	log	checkpoint
MealRecL+	conf	log	checkpoint
iFashion	conf	log	checkpoint

Citation

If you find this work is helpful to your research, please kindly cite the following paper.

@inproceedings{li2025disentangled,
  title={Disentangled Contrastive Bundle Recommendation with Conditional Diffusion},
  author={Li, Jiuqiang},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={11},
  pages={12067--12075},
  year={2025}
}

Acknowledgement

​​This repository is based on CrossCBR, MultiCBR, DiffRec and CoHeat. Thanks for their work.