Humorous Headline Generation via Style Transfer (a.k.a. Humor Translation)

Code and Datasets from the paper, "Can Humor Prediction Datasets be used for Humor Generation? Humorous Headline Generation via Style Transfer" by Orion Weller, Nancy Fulda, and Kevin Seppi.

Related Work

For related projects, see our work on Humor Detection (separating the humorous jokes from the non-humorous) or our collection of 500k+ jokes.

** We do not endorse these jokes. Please view at your own risk **

Overview

Data files are located in data/*.csv and are from the HumicroEdit Dataset. The script to reproduce the dataset files is create_sentences.py. The neural translation model is in transformer.py while the random model comes from the random_translate.py file.

How to Use

Set Up

0. Run pip3 install -r requirements.txt and python -m spacy download en to install the correct packages

Train the Model

0. Run python3 translate.py to start producing translations. If you want to enter in your own sentences interactively after each epoch, use python3 translate.py --interactive

Evaluate Trained Models:

0. Run python3 translate.py -all -e -s to generate a file with outputs
1. Run python3 clean_and_ready_eval.py to add in the Random model and clean up puncuation, etc.

Pre-trained Models

I've included four pre-trained models at various epochs (30, 60, 90, 120). If you'd like to just play with a model, feel free to use one of them (found in the models/{epoch_num}-{timestamp}-model.pt).

Citation

If you found this repository helpful, please cite the following paper:

@inproceedings{weller2020can,
  title={Can Humor Prediction Datasets be used for Humor Generation? Humorous Headline Generation via Style Transfer},
  author={Weller, Orion and Fulda, Nancy and Seppi, Kevin},
  booktitle={Proceedings of the Second Workshop on Figurative Language Processing},
  pages={186--191},
  year={2020}
}