PyTorch version (Default) | CuPy version

This is the official implementation of our ACM The Web Conference 2023 (WWW 2023) paper:

Tianjun Wei, Jianghong Ma, Tommy W.S. Chow. Fine-tuning Partition-aware Item Similarities for Efficient and Scalable Recommendation. [arXiv]

The model implementation ensures compatibility with the Recommendation Toolbox RecBole (Github: Recbole). This is the pure PyTorch version of FPSR model.

The requirements of the running environement:

• Python: 3.8+
• RecBole: 1.2.0
• PyTorch: 1.9.0+

Here we only put zip files of datasets in the respository due to the storage limits. To use the dataset, run

unzip -o "Data/*.zip"

If you like to test FPSR on the custom dataset, please place the dataset files in the following path:

.
|-Data
| |-[CUSTOM_DATASET_NAME]
| | |-[CUSTOM_DATASET_NAME].user
| | |-[CUSTOM_DATASET_NAME].item
| | |-[CUSTOM_DATASET_NAME].inter

And create [CUSTOM_DATASET_NAME].yaml in ./Params with the following content:

dataset: [CUSTOM_DATASET_NAME]

For the format of each dataset file, please refer to RecBole API.

For each dataset, the optimal hyperparameters are stored in Params/[DATASET].yaml. To tune the hyperparamters, modify the corresponding values in the file for each dataset.

The script run.py is used to reproduce the results presented in paper. Train and evaluate FPSR on a specific dataset, run

python run.py --dataset DATASET_NAME

We also provide Colab notebook version of FPSR, you can click here to open Google Colab, select the runtime type as GPU, and run the model.

If you wish, please cite the following paper:

@inproceedings{10.1145/3543507.3583240,
author = {Wei, Tianjun and Ma, Jianghong and Chow, Tommy W. S.},
title = {Fine-Tuning Partition-Aware Item Similarities for Efficient and Scalable Recommendation},
year = {2023},
isbn = {9781450394161},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3543507.3583240},
doi = {10.1145/3543507.3583240},
booktitle = {Proceedings of the ACM Web Conference 2023},
pages = {823–832},
numpages = {10},
keywords = {Recommender System, Similarity Measuring, Collaborative Filtering, Graph Partitioning},
location = {Austin, TX, USA},
series = {WWW '23}
}