Feature Decomposition for Reducing Negative Transfer: a Novel Multi-task Learning Model for Recommender System

Model Architecture

Bibtex:

@article{zhou2023feature,
  title={Feature Decomposition for Reducing Negative Transfer: A Novel Multi-task Learning Method for Recommender System (Student Abstract)},
  author={Zhou, Jie and Yu, Qian and Luo, Chuan and Zhang, Jing},
  year={2023},
}

Project structure

├── code
    ├── aliccp
        ├── FDN.py
        ├── MMoE.py
        ├── PLE.py
        ├── run.sh
    ├── synthetic_dataset
        ├── FDN.py
        ├── MMoE.py
        ├── PLE.py
        ├── run.sh
├── data 
    ├── aliccp
        ├── train_data
            ├── train_data.csv
        ├── test_data
            ├── test_data.csv
        ├── vocab
            ├── ID feature index files
        ├── reader.py
    ├── synthetic_dataset
        ├── train_data
        ├── test_data
        ├── synthesis_dataset_generation.py
├── README.md

Quick start

Ali-CCP: Alibaba Click and Conversion Prediction

cd code/aliccp
sh run.sh

Synthetic DataSet

cd code/synthetic_dataset
sh run.sh

Acknowledgement

• The work is also supported by the National Natural Science Foundation of China (No. 62202025 and 62002012).