File size: 3,869 Bytes
7144397 0aa5e23 0710610 0aa5e23 8c19a5e 0aa5e23 8c19a5e 0aa5e23 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 |
---
license: cc-by-4.0
language:
- en
tags:
- recommender-system
- multimodal
- deep-learning
- academic-paper
datasets:
- Baby
- Sports
- Clothing
metrics:
- recall
- ndcg
---
# 📖 Model Card: [REARM]
**"[Refining Contrastive Learning and Homography Relations for Multi-Modal Recommendation]"**,
*Shouxing Ma, Yawen Zeng, Shiqing Wu, and Guandong Xu*
Published in *[ACM MM]*, 2025.
[[Paper Link](https://arxiv.org/abs/2508.13745)] [[Code Repository](https://huggingface.co/MrShouxingMa/REARM/tree/main)]
---
## ✨ Overview
- We propose a novel multi-modal contrastive recommendation framework (REARM), which preserves recommendation-relevant modal-shared and valuable modal-unique information through meta-network and orthogonal constraint strategies, respectively.
- We jointly incorporate co-occurrence and similarity graphs of users and items, allowing more effective capturing of the underlying structural patterns and semantic (interest) relationships, thereby enhancing recommendation performance.
- Extensive experiments are conducted on three publicly available datasets to evaluate our proposed method. The experimental results show that our proposed framework outperforms several state-of-the-art recommendation baselines.
---
## 🧩 Environment Requirement
The code has been tested running under Python 3.6. The required packages are as follows:
* pytorch == 1.13.0
* numpy == 1.24.4
* scipy == 1.10.1
## Data
Full data could be downloaded from huggingfac:
* [Baby](https://huggingface.co/datasets/MrShouxingMa/Baby/tree/main)
* [Sports](https://huggingface.co/datasets/MrShouxingMa/Sports/tree/main)
* [Clothing](https://huggingface.co/datasets/MrShouxingMa/Clothing/tree/main)
## Dataset
We provide three processed datasets: Baby, Sports, and Clothing.
| #Dataset | #Interactions | #Users|#Items|Sparsity|
| ---- | ---- | ---- |---- |---- |
|Baby|160,792|19,445|7,050|99.88%|
|Sports|296,337|35,598|18,357|99.96%|
|Clothing|278,677|39,387|23,033|99.97%|
## 🚀 Example to Run the Codes
The instructions for the commands are clearly stated in the codes.
* Baby dataset
```
python main.py --dataset='baby' --num_layer=4 --reg_weight=0.0005 --rank=3 --s_drop=0.4 --m_drop=0.6 --u_mm_image_weight=0.2 --i_mm_image_weight=0 --uu_co_weight=0.4 --ii_co_weight=0.2 --user_knn_k=40 --item_knn_k=10 --n_ii_layers=1 --n_uu_layers=1 --cl_tmp=0.6 --cl_loss_weight=5e-6 --diff_loss_weight=5e-5
```
* Sports dataset
```
python main.py --dataset='sports' --num_layer=5 --reg_weight=0.05 --rank=7 --s_drop=1 --m_drop=0.2 --u_mm_image_weight=0 --i_mm_image_weight=0.2 --uu_co_weight=0.9 --ii_co_weight=0.2 --user_knn_k=25 --item_knn_k=5 --n_ii_layers=2 --n_uu_layers=2 --cl_tmp=1.5 --cl_loss_weight=1e-3 --diff_loss_weight=5e-4
```
* Clothing dataset
```
python main.py --dataset='clothing' --num_layer=4 --reg_weight=0.00001 --rank=3 --s_drop=0.4 --m_drop=0.1 --u_mm_image_weight=0.1 --i_mm_image_weight=0.1 --uu_co_weight=0.7 --ii_co_weight=0.1 --user_knn_k=45 --item_knn_k=10 --n_ii_layers=1 --n_uu_layers=1 --cl_tmp=0.03 --cl_loss_weight=1e-6 --diff_loss_weight=1e-5
```
## REARM
The released code consists of the following files.
```
--data
--baby
--clothing
--sports
--utils
--configurator
--data_loader
--evaluator
--helper
--logger
--metrics
--parser
--main
--model
--trainer
```
## Citation
If you want to use our codes and datasets in your research, please cite:
```
@inproceedings{REARM,
title = {Refining Contrastive Learning and Homography Relations for Multi-Modal Recommendation,
author = {Ma, Shouxing and
Zeng, Yawen and
Wu, Shiqing and
Xu, Guandong},
booktitle = {Proceedings of the 33th ACM International Conference on Multimedia},
year = {2025}
}
```
|