MARS

This repository contains a implementation of our "Multi-Facet Recommender Networks with Spherical Optimization" accepted by ICDE 2021.

Contact

Please contact us if you have problems with the code, and also if you think your work is relevant but missing from the survey.

Yanchao Tan (yctan@zju.edu.cn), Xiangyu Wei (weixy@zju.edu.cn)

Environment Setup

1. Pytorch 1.2+
2. Python 3.6+

Guideline

data

We provide a dataset ciao , which contains:

• All interactions of the dataset(ratings.dat);
• Train set, validation set, and test set devided by drop the last two items of each user from ratings.dat(LOOTrain.dat, LOOTest.dat, LOOVal.dat);
• 100 unordered items for each user for testing(LOONegatives.dat);
• 200 pre-sampled data to accelerate the speed of training in 200 epochs(samples/sampling_*.dat);

model

The implementation of MARS(model.py);

utils

Data input and model evaluation

• Dataset.py For Data preprocessing;

• Recommender.py The base class of models, including functions of getting training and testing instances and evaluating performances;

• evaluation.py Functions of calculating NDCG and HR;

Example to run the codes

python main.py --dataset ciao --numEpoch 100 --lRate 0.01

Citation

If you find the code useful, please consider citing the following paper:

@inproceedings{tan2021multi,
  title={Multi-Facet Recommender Networks with Spherical Optimization},
  author={Tan, Yanchao and Yang, Carl and Wei, Xiangyu and Ma, Yun and Zheng, Xiaolin},
  booktitle={2021 IEEE 37th International Conference on Data Engineering (ICDE)},
  pages={1524--1535},
  year={2021},
  organization={IEEE}
}