ADRec: Unlocking the Power of Diffusion Models in Sequential Recommendation ArXiv

An official implementation for the KDD 2025 paper 'Unlocking the Power of Diffusion Models in Sequential Recommendation: A Simple and Effective Approach'.

Jialei Chen, Yuanbo Xu✉ and Yiheng Jiang

Requirements

The following environment packages must be installed to set up the required dependencies.

auto_mix_prep==0.2.0
einops==0.8.0
matplotlib==3.10.0
numpy==2.2.2
PyYAML==6.0.2
scipy==1.15.1
seaborn==0.13.2
torch==2.4.0
torchtune==0.4.0
tqdm==4.66.5

Our code has been tested, running under a Linux server with NVIDIA GeForce RTX 4090 GPU.

Usage

First, navigate to the src directory.

We have provided pre-trained embedding weights, which can be directly used for subsequent backbone warm-up and full-parameter fine-tuning. You can directly run the below command for model training and evaluation.

ADRec:

python main.py --dataset baby --model adrec

Pretrain embedding:

If you want to reproduce the pre-trained weights, you can run the following code:

python main.py --dataset baby --model pretrain

ADRec with multi-task framework PCGrad:

python main.py --dataset baby --model adrec --pcgrad true

We also release some baselines.

DiffuRec:

python main.py --dataset baby --model diffurec

DreamRec:

python main.py --dataset baby --model dreamrec

SASRec+:

python main.py --dataset baby --model sasrec

We also provide a script to run multiple models across various datasets.

bash baseline.bash

t-SNE visualization

The t-SNE visualization experiment can be conducted via /src/t-SNE.ipynb.

Comprehensive evaluation of the original embedding space

A comprehensive evaluation of embedding representations in the original embedding space can be performed using /src/embedding_metrics.ipynb.

Acknowledgements

RecBole, DiffuRec, DreamRec and SASRec+.

📄 Citation

If you find this work useful, please consider citing our paper:

@inproceedings{JLchen2025ADRec,
	title={Unlocking the Power of Diffusion Models in Sequential Recommendation: A Simple and Effective Approach},
	author={Jialei Chen and Yuanbo Xu and Yiheng Jiang},
	booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)},
	year={2025},
	organization={ACM},
	doi = {10.1145/3711896.3737172}
}