Graph Hawkes Neural Network for Forecasting on Temporal Knowledge Graphs

This repository contains code for the reprsentation proposed in Graph Hawkes Neural Network for Forecasting on Temporal Knowledge Graphs paper.

Installation

• Create a conda environment:

$ conda create -n ghnn python=3.6 anaconda
$ source activate ghnn

• Change directory to the GHNN folder
• Install PyTorch (>= 0.4.0)

How to use

After installing the requirements, run the following command to preprocess datasets.:

$ python3 data/DATA_NAME/get_history.py
$ python3 data/DATA_NAME/get_history_tpre_appro.py

To train and test the model.

$ python3 co-train.py -d DATA_NAME  

Only evaluating the model.

$ python3 co-train.py -d DATA_NAME --only_eva true --eva_dir MODEL_DIR

Citation

If you use the codes, please cite the following paper:

@inproceedings{han2020graph,
  title={Graph Hawkes Neural Network for Forecasting on Temporal Knowledge Graphs},
  author={Han, Zhen and Ma, Yunpu and Wang, Yuyi and Günnemann, Stephan and Tresp, Volker},
  booktitle={AKBC},
  year={2020}
}

License

Copyright (c) 2020-present, Siemens AG. All rights reserved. This source code is licensed under the license found in the LICENSE file in the root directory of this source tree.