DiffMM
DiffMM copied to clipboard
[ACM MM'2024]"DiffMM: Multi-Modal Diffusion Model for Recommendation"
DiffMM: Multi-Modal Diffusion Model for Recommendation
This is the PyTorch implementation for DiffMM proposed in the paper DiffMM: Multi-Modal Diffusion Model for Recommendation, which is accepted by ACM MM 2024 Oral.

In this paper, we propose DiffMM, a new multi-modal recommendation model that enriches the probabilistic diffusion paradigm by incorporating modality awareness. Our approach utilizes a multi-modal graph diffusion model to reconstruct a comprehensive user-item graph, while harnessing the advantages of a cross-modal data augmen- tation module that provides valuable self-supervision signals. To assess the effectiveness of DiffMM, we conducted extensive experi- ments, comparing it to several competitive baselines. The results unequivocally establish the superiority of our approach in terms of recommendation performance, firmly establishing its efficacy.
📝 Environment
We develop our codes in the following environment:
- python==3.9.13
- numpy==1.23.1
- torch==1.11.0
- scipy==1.9.1
🎯 Experimental Results
Performance comparison of baselines on different datasets in terms of Recall@20, NDCG@20 and Precision@20:

🚀 How to run the codes
The command lines to train DiffKG on the three datasets are as below. The un-specified hyperparameters in the commands are set as default.
! If you want to run the codes on baby or sports dataset, please firstly follow the instrcution in ./Datasets
- TikTok
python Main.py --data tiktok --reg 1e-4 --ssl_reg 1e-2 --epoch 50 --trans 1 --e_loss 0.1 --cl_method 1
- Baby
python Main.py --data baby --reg 1e-5 --ssl_reg 1e-1 --keepRate 1 --e_loss 0.01
- Sports
python Main.py --data sports --reg 1e-6 --ssl_reg 1e-2 --temp 0.1 --ris_lambda 0.1 --e_loss 0.5 --keepRate 1 --trans 1
👉 Code Structure
.
├── README.md
├── Main.py
├── Model.py
├── Params.py
├── DataHandler.py
├── Utils
│ ├── TimeLogger.py
│ └── Utils.py
├── figures
│ ├── model.png
│ ├── dataset.png
│ └── performance.png
└── Datasets
├── tiktok
│ ├── trnMat.pkl
│ ├── tstMat.pkl
│ ├── valMat.pkl
│ ├── audio_feat.npy
│ ├── image_feat.npy
│ └── text_feat.npy
├── baby
│ ├── trnMat.pkl
│ ├── tstMat.pkl
│ ├── valMat.pkl
│ ├── text_feat.npy
│ └── image_feat.npy.zip
└── README.md
📚 Datasets

🌟 Citation
If you find this work helpful to your research, please kindly consider citing our paper.
@article{jiang2024diffmm,
title={DiffMM: Multi-Modal Diffusion Model for Recommendation},
author={Jiang, Yangqin and Xia, Lianghao and Wei, Wei and Luo, Da and Lin, Kangyi and Huang, Chao},
journal={arXiv preprint arXiv:2406.11781},
year={2024}
}