TaNP

The source code is for our paper: ”Task-adaptive Neural Process for User Cold-Start Recommendation" accepted in WWW 2021.

@inproceedings{lincsr2021,
  title={Task-adaptive Neural Process for User Cold-Start Recommendation},
  author={Lin, Xixun and Wu, Jia and Zhou, Chuan and Pan, Shirui and Cao, Yanan and Wang, Bin},
  booktitle={ACM International World Wide Web Conferences (WWW)},
  year={2021}
}

Requirements

python == 3.6.3
torch == 1.1
numpy == 1.17
scipy == 1.3.1
scikit-learn == 0.21.3

Datasets

MoviesLens-1M is provided by MeLU, and you can find it from here.
Last.FM is provided by MKR, and you can find it from here.
Gowalla is provided by NGCF, and you can find it from here.

How to use

The utils/loader is used for data preprocessing, and you can customize this part for your own data.
Use the following commands for running a sub-dataset:
unzip data.zip file
zsh train.sh # Training with default hyper-parameters.