FactOwl: Blazingly fast and modern factchecker

Usage Examples

1. Fact generation and verification using pre-computed contexts only on RiDiC dataset

2. Fact generation and verification on FactScore dataset usign either fixed Wikipedia dump or Wikipedia search for context retrieval

3. LLM and rule-based fact filtration

Installation

For development:

conda update --file factowl.yml
conda activate ltf
pip install -e ./factowl/
python -m spacy download en_core_web_sm
python factowl/factowl/download_data.py

and also download enwiki dump from this link and place it into ./data/ directory.

For usage:

pip install factowl

Usage

To run the scorer on the provided sample dataset, simply use:

python -m factowl --input_file data/sample_input.json

You can also provide your own data into the script in the following format:

{
  "topics": [
    "{topic_id}": "{topic_name}"
  ],
  "generations": [
    "{topic_id}": "{generation}"
  ]
}