This repository contains the source code of our paper, Text Generation from Knowledge Graphs with Graph Transformers, which is accepted for publication at NAACL 2019.
Training:
python3.6 train.py -save <DIR>
Use --help for a list of all training options.
To generate, use
python3.6 generator.py -save <SAVED MODEL>
with the appropriate model flags used to train the model
To evaluate, run
python3.6 eval.py <GENERATED TEXTS> <GOLD TARGETS>
The AGENDA dataset is available in a user-friendly json format in /data/unprocessed.tar.gz Preprocessed data is also available in /data.
If this work is useful in your research, please cite our paper.
@inproceedings{koncel2019text,
title={{T}ext {G}eneration from {K}nowledge {G}raphs with {G}raph {T}ransformers},
author={Rik Koncel-Kedziorski, Dhanush Bekal, Yi Luan, Mirella Lapata, and Hannaneh Hajishirzi},
booktitle={NAACL},
year={2019}
}