HGLS

This repository contains the code for the ACM Web Conference (WWW') 2023 paper titled "Learning Long- and short-term representations for Temporal Knowledge Graph Reasoning".

Usage

Generate data

You need to run the file generate_data.py to generate the graph data needed for our model:

python generate_data.py --data=DATA_NAME

In order to speed up training and testing, for ICEWS18, ICEWS05-15, and GDELT datasets, data in the required format can be constructed in advance before training and testing:

python save_data.py --data=DATA_NAME

Training and Testing

Then you can run the file main.py to train and test our model. The detailed commands can be found in {dataset}.sh. Some important hyper-parameters can be found in long_config.yaml and short_config.yaml.

Requirements

Make sure you have the following dependencies installed:

• Python~=3.7
• dgl~=0.9.1
• torch~=1.12.1
• numpy~=1.21.5
• tqdm~=4.64.1
• pandas~=1.3.5
• scipy~=1.7.3

Citation

Please cite our paper if you use the code:

@inproceedings{zhang2023learning,
  title={Learning Long-and Short-term Representations for Temporal Knowledge Graph Reasoning},
  author={Zhang, Mengqi and Xia, Yuwei and Liu, Qiang and Wu, Shu and Wang, Liang},
  booktitle={Proceedings of the ACM Web Conference 2023},
  pages={2412--2422},
  year={2023}
}

Acknowledge

Some of our code is also referenced from RE-GCN: https://github.com/Lee-zix/RE-GCN.