BIGCF

This is the PyTorch implementation for our SIGIR 2024 paper:

Yi Zhang, Lei Sang, and Yiwen Zhang*. 2024. Exploring the Individuality and Collectivity of Intents behind Interactions for Graph Collaborative Filtering. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’24).

Environment

python == 3.8.18
pytorch == 2.1.0 (cuda:12.1)
torch-sparse == 0.6.18
scipy == 1.10.1
numpy == 1.24.3

Examples to run the codes

We adopt three widely used large-scale recommendation datasets: Gowalla, Amazon-Book, and Tmall. BIGCF is an easy-to-use recommendation model in which the most important hyperparameter is the weight of the contrastive loss ssl_reg, and the other parameters can be set by default. The following are examples of runs on three datasets:

• Gowalla:

  python main.py --dataset gowalla --epoch 150 --ssl_reg 0.4

• Amazon-Book:

  python main.py --dataset amazon --epoch 200 --ssl_reg 0.4

• Tmall:

  python main.py --dataset tmall --epoch 200 --ssl_reg 0.2

The log folder provides training logs for reference. The results of a single experiment may differ slightly from those given in the paper because they were run several times and averaged in the experiment.

Different Frameworks

Note that the experiments in the paper on BIGCF take the same training framework as previous work (DCCF). In the meantime, we also provide the code for BIGCF in the same training framework as the traditional classical work (e.g., LightGCN), see: >[https://github.com/BlueGhostYi/BIGCF-full-sample]

Citation

If you find this work is helpful to your research, please consider citing our paper:

@inproceedings{zhang2024exploring,
  title={Exploring the Individuality and Collectivity of Intents behind Interactions for Graph Collaborative Filtering},
  author={Zhang, Yi and Sang, Lei and Zhang, Yiwen},
  booktitle={Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
  pages={1253--1262},
  year={2024}
}

Acknowledgement

To maintain fair comparisons and consistency, our codes are mainly adapted from the following repo:

[https://github.com/HKUDS/DCCF]

Many thanks to them for providing the training framework and for the active contribution to the open source community.