yule-BUAA / SFCNTSP

codes of Predicting Temporal Sets with Simplified Fully Connected Networks at AAAI 2023

Predicting Temporal Sets with Simplified Fully Connected Networks

The description of "Predicting Temporal Sets with Simplified Fully Connected Networks" at AAAI 2023 is available here.

Original data:

The original data could be downloaded from here. You can download the data and then put the data files in the ./original_data folder.

To run the code:

1. run ./preprocess_data/preprocess_data_{dataset_name}.py to preprocess the original data, where dataset_name could be JingDong, DC, TaoBao and TMS. We also provide the preprocessed datasets at here, which should be put in the ./dataset folder.

2. run ./train/train_SFCNTSP.py to train the model and get the results on different datasets according to the configuration in ./utils/config.json.

Environments:

• PyTorch 1.8.1
• tqdm
• numpy
• pandas

Hyperparameter settings:

Hyperparameters can be found in ./utils/config.json file, and you can adjust them when training the model on different datasets.

Hyperparameters	JingDong	DC	TaoBao	TMS
learning rate	0.001	0.001	0.001	0.001
dropout rate	0.2	0.1	0.05	0.1
embedding channels	64	64	32	64
alpha	1.0	1.0	1.0	1.0
beta	0.1	0.1	0.1	0.1

Citation

Please consider citing our paper when using the codes or datasets.

@inproceedings{yu2023predicting,
  title={Predicting Temporal Sets with Simplified Fully Connected Networks},
  author={Yu, Le and Liu, Zihang and Zhu, Tongyu and Sun, Leilei and Du, Bowen and Lv, Weifeng},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={37},
  number={4},
  pages={4835--4844},
  year={2023}
}