Modality-Balanced Learning for Multimedia Recommendation

This is the code for the ACM Multimedia 2021 Paper: Modality-Balanced Learning for Multimedia Recommendation.

Requirements

• Python 3.8
• PyTorch 1.9.1

Dataset

Please refer to this link for the dataset preparation.

Quick Start

Start training and inference as:

cd codes
# train textual teacher model
python train_unimodal.py --dataset baby  --model_name VBPR --train_type 2 --save_model 1
# train visual teacher model
python train_unimodal.py --dataset baby  --model_name VBPR --train_type 3 --save_model 1
# train multimodal student model
python train_kd.py --dataset baby  --model_name VBPR  

Citation

Please cite our paper if you use the code:

@inproceedings{10.1145/3664647.3680626,
author = {Zhang, Jinghao and Liu, Guofan and Liu, Qiang and Wu, Shu and Wang, Liang},
title = {Modality-Balanced Learning for Multimedia Recommendation},
year = {2024},
url = {https://doi.org/10.1145/3664647.3680626},
doi = {10.1145/3664647.3680626},
booktitle = {Proceedings of the 32nd ACM International Conference on Multimedia},
pages = {7551–7560},
numpages = {10},
}