This is the implementation of EMNLP 2021 paper Injecting Entity Types into Entity-Guided Text Generation.

In this work, we aim to enhance the role of entity in NLG model to help generate sequence accurately. Specifically, we develop a novel NLG model to produce a target sequence based on a given list of entities. Our model has a multi-step decoder that injects the entity types into the process of entity mention generation. Experiments on Gigawords and NYT demonstrate type injection performs better than existing type concatenation baselines.

1. Using pip: The first way is to install the required packages in requirements.txt file by using pip. Create the environment by running the following command: pip install -r requirements.txt

2. Using docker: The second way is to pull our docker image on the dockerhub. Download the docker image by using the following command: docker pull wenhaoyu97/injtype:gen1.0

To prepare the dataset, run python dataset/dataloader.py in the top folder directory.

To run the model with Gigawords Dataset
(1) NQG_HOME=/home_dir_replace_with_yours/InjType
(2) bash $NQG_HOME/scripts/inj_gig.sh $NQG_HOME

To run the model with NYT Dataset
(1) NQG_HOME=/home_dir_replace_with_yours/InjType
(2) bash $NQG_HOME/scripts/inj_nyt.sh $NQG_HOME

We use Texar-torch BLEU score and PyPI ROUGE to evaluate model performance.

@inproceedings{dong2021injecting,
  title={Injecting Entity Types into Entity-Guided Text Generation},
  author={Dong, Xiangyu and Yu, Wenhao and Zhu, Chenguang and Jiang, Meng},
  booktitle={Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year={2021}
}

This code was based in part on the source code of NQG.

If you have any question or suggestion, please send email to:
Xiangyu Dong ([email protected]) or Wenhao Yu ([email protected])