⚡ LeapRec

This project is a PyTorch implementation of "Calibration-Disentangled Learning and Relevance-Prioritized Reranking for Calibrated Sequential Recommendation" (LeapRec) which is published at CIKM 2024.

The implementation is based on Python 3.10.0 and PyTorch 2.1.2. A complete list of required packages can be found in the requirements.txt file. Please install the necessary packages before running the code.

💾 Datasets

We use four datasets in our work: ML-1M, Goodreads, Grocery, and Steam. The preprocessed dataset is included in the repository: ./data.

💻 Source codes

Source codes are included in ./src. The codes are divided into two repositories, backbone and reranking. They are respectively for training a backbone model and reranking the results.

⚙️ Pretrained backbone models

In ./out, we provide pretrained backbone models which are trained under our proposed calibration-aware learning to rank. Therefore, you can run reranking algorithm without training the model.

🚀 Running the code

To run the model training code, use the python main.py with an argument --data in the repository ./backbone. To run the reranking code, use the python main.py with arguments --data, --algorithm, and --balance in the repository ./reranking. The argument --balance is lambda of our paper. For convenience, we provide a demo.sh script that reproduces the experiments presented in our paper.

To customize the configuration, please edit the config.yaml file in each repository. Please refer to our paper if you need guidance on setting the hyperparameters.

📚 Citation

To cite LeapRec in your work, please use the following BibTeX entry:

@inproceedings{jeon24calibration,
  title = "Calibration-disentangled learning and relevance-prioritized reranking for calibrated sequential recommendation",
  author = "Hyunsik Jeon and Se-eun Yoon and Julian McAuley",
  year = "2024",
  booktitle = "CIKM"
}