Re4: Learning to Re-contrast, Re-attend, Re-construct for Multi-interest Recommendation

Re4: Learning to Re-contrast, Re-attend, Re-construct for Multi-interest Recommendation
Shengyu Zhang, Lingxiao Yang, Dong Yao and Yujie Lu, Fuli Feng, Zhou Zhao, Tat-Seng Chua, Fei Wu
The ACM Web Conference 2022 (WWW 2022)
Key Words:  Recommender Systems;  Multi-interest;  Backward Flow
[Paper], [Slides]

A pytorch implementation of Re4

Prerequisites

• Python 3
• PyTorch 1.8.1
• TensorFlow 2.x

Getting Started

Installation

• Install PyTorch 1.8.1
• Install TensorFlow 2.x
• Clone this repository git clone https://github.com/DeerSheep0314/Re4-Learning-to-Re-contrast-Re-attend-Re-construct-for-Multi-interest-Recommendation.git.

Dataset

• Amazon-book dataset can be downloaded through:
  • Microsoft OneDrive [Link]

Running

To run the code, You can use python src/model.py --gpu {gpu_num} --thre {thre_num} --data {dataset_name} --ct_lambda {ct_weight} --cs_lambda {cs_weight} --att_lambda {att_weight} --numin {num_interests} to train the R4 model on a specific dataset. You can set the above hyperparameters here, see the code for other hyperparameters.

For example, you can use python src/model.py --gpu 0 --thre -1 --numin 8 --data book --ct_lambda 0.1 --cs_lambda 0.1 --att_lambda 0.001 to train R4 model on Amazon-book dataset.

Bibtex

@inproceedings{DBLP:conf/www/ZhangYYLFZC022,
  author    = {Shengyu Zhang and
               Lingxiao Yang and
               Dong Yao and
               Yujie Lu and
               Fuli Feng and
               Zhou Zhao and
               Tat{-}Seng Chua and
               Fei Wu},
  title     = {Re4: Learning to Re-contrast, Re-attend, Re-construct for Multi-interest
               Recommendation},
  booktitle = {{WWW} '22: The {ACM} Web Conference 2022},
  pages     = {2216--2226},
  publisher = {{ACM}},
  year      = {2022},
  url       = {https://doi.org/10.1145/3485447.3512094},
}