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