Weak Reward Model Transforms Generative Models into Robust Causal Event Extraction Systems

1. Setup a virtual environment
2. Install dependencies with ./setup Requires uv or pip-tools.

I recommend using uv for managing dependencies because it's a lot faster, but it should work with python's built-in venv and pip-tools as well.

Each project is a separate directory with its own README.md file. They all use PDM to manage dependencies, but we use pip-tools for a repository-wide environment.

• Human Evaluation:
  • agreement: Calculate agreement between LLM judges and human evaluation
  • chatgpt: Use GPT OpenAI API to extract causal events
• Baseline:
  • sequence_labelling: BIO labelling-based model for causal event extraction
  • extractive_qa: Span-based model for causal event extraction
  • gen_qa: QA-based model for causal event extraction
• Our RL framework
  • data: Datasets for the project, including processed data
  • preprocess: Scripts to preprocess data for the different models
  • self_critique: LLM-based extraction, supervised and RL training
  • error_analysis: Analyze errors in the extraction LLM model

If you find our work useful, please cite as:

@misc{silva2024weak,
    title={Weak Reward Model Transforms Generative Models into Robust Causal Event Extraction Systems},
    author={Italo Luis da Silva and Hanqi Yan and Lin Gui and Yulan He},
    year={2024},
    eprint={2406.18245},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

This project is licensed under the GPL version 3 or later.