This repository hosts the codefor the paper "UIPC-MF: User-Item Prototype Connection Matrix Factorization for Explainable Collaborative Filterings" by Lei Pan and Von-Wun Soo.

@article{pan2023uipc,
  title={UIPC-MF: User-Item Prototype Connection Matrix Factorization for Explainable Collaborative Filtering},
  author={Pan, Lei and Soo, Von-Wun},
  journal={arXiv preprint arXiv:2308.07048},
  year={2023}
}

The code is derived from https://github.com/karapostK/ProtoMF. The structure of the code is the same as https://github.com/karapostK/ProtoMF.

The code is written in Python and relies on Pytorch to compute the model's gradients. We further use ray tune for hyperparameter optimization and weight and biases for logging, among many cool packages.

• Install the environment with conda env create -f uipc_mf.yml
• Activate the environment with conda activate uipc_mf

All preprocessed data is in /data.

In utilities/consts.py you need to set:

• DATA_PATH: absolute path to the ./data folder
• WANDB_API_KEY: API key of Weight and Biases, currently the results are only printed there.

Running start.py will initiate the training process.

By default, python start.py runs the hyperparameter optimization for a single seed (check utilities/consts.py).

Results and progress can be checked on the Weight&Biases webpage.

The code in this repository is licensed under the Apache 2.0 License.