ScenarioMeta

project | arXiv

Sequential Scenario-Specific Meta Learner for Online Recommendation

Zhengxiao Du, Xiaowei Wang, Hongxia Yang, Jingren Zhou, Jie Tang

Accepted to KDD 2019 Applied Data Science Track!

Under construction. Expect a stable release of cleaner code in June

Prerequisites

Python 3
PyTorch >= 1.0.0
NVIDIA GPU + CUDA cuDNN

Getting Started

Installation

Clone this repo.

git clone https://github.com/THUDM/ScenarioMeta
cd ScenarioMeta

Please install dependencies by

pip install -r requirements.txt

Dataset

Three public datasets are used for experiments. The Taobao Cloud Theme Click Dataset is released by us.

Amazon Review Source
Movielens-20M Source
Taobao Cloud Theme Click Dataset Source

You can download the preprocessed datasets from the link in OneDrive or by running python scripts/download_preprocessed_data.py.
If you're in regions where OneDrive is not available (e.g. Mainland China), try to download from Tsinghua Cloud by running python scripts/download_preprocessed_data.py --tsinghua.

Training

For training, simply run python src/main.py with necessary parameters.

Different configurations for datasets in the paper are stored under the configs/ directory. Launch a experiment with --config to specify the configuration file, --root_directory to specify the path to the preprocessed data, --comment to specify the experiment name which will be used in logging and --gpu to speficy the gpu id to use.

Cite

Please cite our paper if you use the code or datasets in your own work:

@article{du2019scenariometa,
  title={Sequential Scenario-Specific Meta Learner for Online Recommendation},
  author={Du, Zhengxiao and Wang, Xiaowei and Yang, Hongxia and Zhou, Jingren and Tang, Jie},
  journal={arXiv preprint arXiv:1906.00391},
  year={2019}
}