Source code and data for Counterfactual Recipe Generation: Exploring Models’ Compositional Generalization Ability in a Realistic Scenario (EMNLP2022 main conference paper)

• Python>=3.7
• nltk
• bert_score
• ltp
• sklearn

Our data is in the data/ folder. Task data:

• dish_pairs.txt: 50 pairs of (base dish, target dish)
• changing_ingres.txt: 50 pairs of (original ingredient, changed ingredient)
• base_recipes.txt: 2500 base recipes, each line is in the format of base dish \t target dish \t base recipe.
• recipe_corpus_finetune.zip: 1,479,764 recipes of other dishes for model training. Link: https://drive.google.com/file/d/16_X3I3mt-eIaVx5g1ZxA0uks_fD_IBiy/view?usp=sharing

L2 evaluation data:

• glossary_dict.pkl: glossary of ingredient classes, verb classes, and tool classes
• parsing_data.pkl: data used in parsing recipes into actions
• pivot_actions.pkl: pivot actions and order constraints

The code is in the code/ folder. Please change EVAL_TEXT_PATH to the path of generated recipes, and WORD_EMBEDDING_PATH to the word embedding path.

L1 evaluation (coverage of ingredients and extent of preservation):

python L1_eval.py

L2 evaluation (action-level):

python L2_eval.py

Please cite our paper if this repository inspires your work.

@article{liu2022counterfactual,
  title={Counterfactual Recipe Generation: Exploring Compositional Generalization in a Realistic Scenario},
  author={Liu, Xiao and Feng, Yansong and Tang, Jizhi and Hu, Chengang and Zhao, Dongyan},
  journal={arXiv preprint arXiv:2210.11431},
  year={2022}
}

If you have any questions regarding the code, please create an issue or contact the owner of this repository.