COHESION: Composite Graph Convolutional Network with Dual-Stage Fusion for Multimodal Recommendation

Introduction

This is the Pytorch implementation for our COHESION paper:

COHESION: Composite Graph Convolutional Network with Dual-Stage Fusion for Multimodal Recommendation

Environment Requirement

• python 3.9
• Pytorch 2.1.0

Dataset

All experiments conducted on three processed datasets: Baby, Sports, Clothing

Training

cd ./src
python main.py

Performance Comparison

Citing COHESION

If you find COHESION useful in your research, please consider citing our paper.

@inproceedings{xu2025cohesion,
  title={COHESION: Composite Graph Convolutional Network with Dual-Stage Fusion for Multimodal Recommendation},
  author={Xu, Jinfeng and Chen, Zheyu and Wang, Wei and Hu, Xiping and Kim, Sang-Wook and Ngai, Edith CH},
  booktitle={Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages={1830--1839},
  year={2025}
}

Acknowledgement

The structure of this code is based on MMRec. Thank for their work.