😆 → Getting Serious about Humor with LLMs → 😐

Code for Getting Serious about Humor: Crafting Humor Datasets with Unfunny Large Language Models (ACL 2024).

Structure

• data_generation
  Contains logic for generating synthetic data and the corresponding outputs.

• datasets
  Includes original human data for both the Unfun and English-Hindi tweets humor datasets.

• humor_detection
  Houses the logic for training humor classifiers using both synthetic and human humor data.

• evaluation
  Contains human evaluation results, automatic evaluation metrics, and relevant scripts.

• utils
  Utilities for interacting with LLM APIs and handling various file I/O operations.

Getting Started

1. Set up a Python environment:
   Create and activate a Python environment using conda or venv. Python 3.9 is recommended.

2. Install dependencies:
   After activating your environment, install the required packages by running:

   pip install -r requirements.txt

TODO

• Trim unnecessary packages from requirements.txt.
• Include/Improve readability of:
  • Data preprocessing scripts
  • English-Hindi data generation scripts

Citation

@misc{horvitz2024gettinghumorcraftinghumor,
      title={Getting Serious about Humor: Crafting Humor Datasets with Unfunny Large Language Models}, 
      author={Zachary Horvitz and Jingru Chen and Rahul Aditya and Harshvardhan Srivastava and Robert West and Zhou Yu and Kathleen McKeown},
      year={2024},
      eprint={2403.00794},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2403.00794}, 
}