Text Matching Improves Sequential Recommendation by Reducing Popularity Biases

Source code for our CIKM 2023 paper :
Text Matching Improves Sequential Recommendation by Reducing Popularity Biases

Click the links below to view our papers, checkpoints, datasets

If you find this work useful, please cite our paper and give us a shining star 🌟

@inproceedings{liu2023text,
  title={Text Matching Improves Sequential Recommendation by Reducing Popularity Biases},
  author={Liu, Zhenghao and Mei, Sen and Xiong, Chenyan and Li, Xiaohua and Yu, Shi and Liu, Zhiyuan and Gu, Yu and Yu, Ge},
  booktitle={Proceedings of the 32nd ACM International Conference on Information and Knowledge Management},
  pages={1534--1544},
  year={2023}
}

We completed our oral presentation in Birmingham, UK 😊
Our slide show can be viewed by clicking on this link

Requirements

1. Install the following packages using Pip or Conda under this environment

python>=3.8
transformers==4.22.2
numpy==1.23.5
datasets==2.11.0
faiss-cpu
scikit-learn>=1.1.2
pandas
tensorboard

We provide the version file reproduce/environment.yml of all our used packages, if you have any problems configuring the environment, please refer to this document.

Please note! We find that when using a higher version of Transformer, the evaluation will report an error. Please use versions below 4.24.0.😢

2. Install openmatch. To download OpenMatch as a library and obtain openmatch-thunlp-0.0.1.

git clone https://github.com/OpenMatch/OpenMatch.git
cd OpenMatch
pip install -e.

We do not include all the requirements in the package. You may need to manually install torch

3. Prepare the pretrained T5 weights

TASTE is built on t5-base model，

You can download the weights from this link and place them in TASTE/pretrained_model/

Reproduce TASTE

Download Code & Dataset

Congratulations, you've done all the prep work. Let's get started !

First, use git clone to download this project:

git clone https://github.com/OpenMatch/TASTE
cd TASTE

Next, download our processed dataset from the Google Drive, please make sure that the files under the data folder contain the following before running::

data/
├── beauty/
│   ├── train.txt
│   ├── valid.txt
│   ├── test.txt
│   ├── item.txt
│   ├── train_name.jsonl
│   ├── valid_name.jsonl
│   └── item_name.jsonl
├── toys/
├── sports/  
└── yelp/

We have also provided a script for processing the data in the TASTE/data, so you can download the original dataset and use the script to process it yourself.

Train then test

Next, start training the TASTE model In the reproduce folder, we provide scripts for reproducing all TASTE models. Using the beauty dataset as an example, I will show you how to reproduce the results in the TASTE paper.

0⃣️ First.

cd TASTE

1⃣️ Training TASTE on the beauty dataset

bash reproduce/train/beauty/train_beauty.sh

2⃣️ Validate for the best checkpoint

bash reproduce/valid/beauty/eval_beauty.sh

3⃣️ Test for final results

bash reproduce/test/beauty/test_beauty.sh

Acknowledgement

• OpenMatch This repository is built upon OpenMatch! An all-in-one toolkit for information retrieval!
• FID The model architecture of TASTE refers to FID!
• RecBole Data processing and evaluation code based on RecBole! RecBole is developed based on Python and PyTorch for reproducing and developing recommendation algorithms in a unified, comprehensive and efficient framework for research purpose!

Contact

If you have questions, suggestions, and bug reports, please email:

[email protected]