Code and data for the paper "HAUSER: Towards Holistic and Automatic Evaluation of Simile Generation" (ACL 2023)
conda create -n hauser python=3.6.13
conda activate hauser
pip install -r requirements.txt
Install the necessary dependencies for simile component extraction.
cd HAUSER/dependency
bash install.sh
python test_syntax.py
If test_syntax.py runs successfully, it proves that the dependency installation is successful.
Cache the necessary pre-trained models
cd HAUSER/dependency
python test_models.py
Download the Million-scale Simile Knowledge Base MAPS-KB from Google Drive.
Put the file MAPS-KB.csv into the folder [HAUSER/dependency].
First, we finetune a pre-trained sequence-to-sequence model, BART, based on the simile generation datasets from MAPS-KB.
Then, we generate five simile candidates for 2500 literal sentences in the test set.
The data is in HAUSER/data/simile_candidates_raw.json.
You can get the relevance_score, logic_consistency_score, sentiment_consistency_score, creativity_score and informativeness_scoreusing the following script score.sh:
cd HAUSER
python code/score.py
@article{he2023hauser,
title={HAUSER: Towards Holistic and Automatic Evaluation of Simile Generation},
author={He, Qianyu and Zhang, Yikai and Liang, Jiaqing and Huang, Yuncheng and Xiao, Yanghua and Chen, Yunwen},
journal={arXiv preprint arXiv:2306.07554},
year={2023}
}