Skip to content

Latest commit

 

History

History
236 lines (168 loc) · 9.56 KB

README.md

File metadata and controls

236 lines (168 loc) · 9.56 KB

PARA

This is the PyTorch implementation of our paper:

Plug-and-play Rating Prediction Adjustment through Trisecting-acting-outcome

The Workflow of PARA

Requirements

  • numpy=1.21.5
  • pandas=1.3.5
  • python=3.7.16
  • pytorch=1.13.1

Datasets and Pre-trained Models

Dataset information

Dataset Users Items Interactions Link
Amazon-Music 478 235 266 414 836 006 URL
Ciao 17 615 16 121 72 665 URL
Douban-Book 46 548 212 995 1 908 081 URL
Douban-Movie 94 890 81 906 11 742 260 URL
MovieLens-1M 6 040 3 900 1 000 209 URL
MovieLens-10M 69 878 10 677 10 000 054 URL

Pre-trained model information

Method MF-based LightGCN-based ItemCF-based
MF ✅
LightGCN ✅
ItemCF ✅
IPS ✅
DICE ✅
PDA ✅ ✅
TIDE ✅ ✅
PARA ✅ ✅ ✅

Simply Reproduce the Results

Take the Ciao dataset as an example:

  1. Clone the source code

    git clone https://github.com/A-Egoist/TWDP.git --depth=1

  2. Download preprocessed data and pre-trained models Download the data and models from Baidu Netdisk.

    • Each original dataset is split into 5 sets for five-fold cross-validation, consisting of *.train, *.test, and *.extend files:

      • *.train: used for training.
      • *.test: used for testing.
      • *.extend: includes (user, positiveItem, negativeItem) data generate by performing negative sampling on the *.train file.
      • Additionally, the *.pkl files store the processed results of the *.train files.

    • The pre-trained model files follow the naming convention backbone-method-dataset-fold_index. For example, for the PARA method with MF as the backbone model on the Ciao dataset, the models are named as:

      • MF-PARA-Ciao-1.pt
      • MF-PARA-Ciao-2.pt
      • MF-PARA-Ciao-3.pt
      • MF-PARA-Ciao-4.pt
      • MF-PARA-Ciao-5.pt

      These files correspond to models trained on different folds for the five-fold cross-validation experiment.

    • To reproduce the results of PARA method with MF as the backbone model on the Ciao dataset, ensure the dataset contains the following files:

      • movie-ratings1.test, movie-ratings1.pkl
      • movie-ratings2.test, movie-ratings2.pkl
      • movie-ratings3.test, movie-ratings3.pkl
      • movie-ratings4.test, movie-ratings4.pkl
      • movie-ratings5.test, movie-ratings5.pkl

    • Additionally, download the corresponding pre-trained models:

      • MF-PARA-ciao-1.pt
      • MF-PARA-ciao-2.pt
      • MF-PARA-ciao-3.pt
      • MF-PARA-ciao-4.pt
      • MF-PARA-ciao-5.pt

  3. Confirm the file structure Verify the downloaded files and folder structure according to the provided tree.txt file.

  4. Run inference To evaluate the model, run the following command:

    # Windows
python .\main.py --backbone MF --method PARA --dataset ciao --mode eval

    Avaliable Options:

    • --backbone: The backbone model. Available options: ['MF', 'LightGCN'].
    • --method: The method to be used. Available options: ['Base', 'IPS', 'DICE', 'PDA', 'TIDE', 'PARA'].
    • --dataset: The dataset to use. Available options: ['amzoun-music', 'ciao', 'douban-book', 'douban-movie', 'ml-1m', 'ml-10m'].
    • --mode: The mode to be choosen. Available options: ['train', 'eval', 'both'].

  5. Convert the results to Excel After evaluation, convert the log results into an Excel file using the following command:

    # Windows
python .\logs\log_to_excel.py --input .\logs\eval.py --output .\output\eval.xlsx --sl 1 --el 2000

    Explanation of Parameters:

    • --input: Specifies the log file that contains the evaluation results (e.g., eval.py).
    • --output: Specifies the base name of the output Excel file (e.g., eval.xlsx). The actual output files will be saved as three separate files:
      • eval-mean.xlsx: Contains the mean of the evaluation metrics.
      • eval-std.xlsx: Contains the standard deviation of the evaluation metrics.
      • eval-rank.xlsx: Contains the ranking of the evaluation metrics.
    • --sl: Specifies the start line in the log file from which the results will be converted to Excel.
    • --el: Specifies the end line in the log file up to which the results will be converted to Excel.

Start from Scratch

This section explains how to reproduce the results from scratch, taking the Ciao dataset as an example:

  1. Clone source code and datasets Clone the repository containing the code:

    git clone https://github.com/A-Egoist/TWDP.git --depth=1

    Download the Ciao dataset from URL and move it into the corresponding folder according to tree.txt

  2. Data preprocessing

    (a). Split the dataset into 5 sets

    Run the following command to split the Ciao dataset into 5 subsets for five-fold cross-validation:

    # Windows
python .\src\data_processing.py --dataset ciao

    Available Option:

    • --dataset: The dataset to be splited. Available options: ['amzoun-music', 'ciao', 'douban-book', 'douban-movie', 'ml-1m', 'ml-10m'].

    (b). Compile the negative sampling script

    Use the following command to compile the C++ script for negative sampling:

    # Windows
g++ .\src\negative_sampling.cpp -o .\src\negative_sampling.exe

    (c). Perform negative sampling

    Execute the compiled script to perform negative sampling:

    # Windows
.\src\negative_sampling.exe ciao 1

    Explanation of Parameters:

    • The first parameter specifies the dataset to be processed, with available options: ['amazon-music', 'ciao', 'douban-book', 'douban-movie', 'ml-1m', 'ml-10m']
    • The second parameter specifies the fold index for the cross-validation dataset, with options: ['1', '2', '3', '4', '5']

  3. Training To start the training process, run the following command:

    # Windows
python .\main.py --backbone MF --method PARA --dataset ciao --mode train

    Available Options:

    • --backbone: The backbone model. Available options: ['MF', 'LightGCN'].
    • --method: The method to be used. Available options: ['Base', 'IPS', 'DICE', 'PDA', 'TIDE', 'PARA'].
    • --dataset: The dataset to use. Available options: ['amzoun-music', 'ciao', 'douban-book', 'douban-movie', 'ml-1m', 'ml-10m'].
    • --mode: The mode to be chosen. Available options: ['train', 'eval', 'both'].

  4. Evaluation After training, evaluate the model's performance using this command:

    # Windows
python .\main.py --backbone MF --method PARA --dataset ciao --mode eval

    Options are the same as step 3.

  5. Convert the results to Excel Finally, convert the evaluation logs to an Excel file:

    # Windows
python .\logs\log_to_excel.py --input .\logs\eval.py --output .\output\eval.xlsx --sl 1 --el 2000

    Explanation of Parameters:

    • --input: Specifies the log file that contains the evaluation results (e.g., eval.py).
    • --output: Specifies the base name of the output Excel file (e.g., eval.xlsx). The actual output files will be saved as three separate files:
      • eval-mean.xlsx: Contains the mean of the evaluation metrics.
      • eval-std.xlsx: Contains the standard deviation of the evaluation metrics.
      • eval-rank.xlsx: Contains the ranking of the evaluation metrics.
    • --sl: Specifies the start line in the log file from which the results will be converted to Excel.
    • --el: Specifies the end line in the log file up to which the results will be converted to Excel.

Citation

If you use this code, please cite the following paper:

@article{ZhangLong2024PARA,
  title   = {Plug-and-play Rating Prediction Adjustment through Trisecting-acting-outcome},
  author  = {},
  journal = {},
  year    = {},
  volume  = {},
  number  = {},
  pages   = {},
  doi     = {}
}

Acknowledgments