/codes.crossdomain.rec

Primary LanguageC++

CPR: Cross-domain Preference Ranking with User Transformation

Our codes for https://dl.acm.org/doi/abs/10.1007/978-3-031-28238-6_35

0. Environment

  • Docker image: nvcr.io/nvidia/pytorch:22.05-py3
pip install -r requirements.txt

1. Data

  • Enter preprocess directory
cd preprocess/
  • Download raw data
bash download_amazon_data.sh {raw_data_dir}
  • Process raw to loo (It takes long time, particularly Books_5)
bash raw_to_loo.sh {raw_data_dir} {loo_data_dir}
  • Process loo to loo-5core
bash loo_to_ncore.sh {loo_data_dir} {ncore_data_dir}
  • Generate input for CPR
bash generate_cpr_input.sh {ncore_data_dir} {cpr_input_dir}

2. Model Training & Evaluation

  • Enter CPR directory
$ cd models/CPR
  • Compile cpp code
make
  • Usage example
bash run_cpr.sh\
        {ncore_data_dir}\
        {cpr_input_dir}\
        {emb_save_dir}\
        {exp_record_dir}\
        traineval\
        hk\
        csjj