FACTS: Table Summarization via Offline Template Generation with Agentic Workflows

Folder Structure

|- conf/  # Configuration files
    |- config.yaml  # Global configuration
|- data/  # Data files
    |- readme.md  # the original sources of each dataset we used
|- src/  # Source code for the project
|- .gitignore  # Git ignore file
|- main.py  # Main file to run the project
|- pyproject.toml  # Config isort and black to better format the code
|- README.md  # This file
|- requirements.txt  # necessary dependencies for pip install
|- utils.py  # Utility functions for the entire project

Create Conda Environment

We can automaticaly create a conda environment through:

conda create -n facts python=3.13
conda activate facts
pip install -r requirements.txt

Run with our FACTS framework

We can run experiments on different datasets by modifying the config.yaml file.

python /user_directory/table_summarization/main.py mode=facts_framework

Citation

If you find FACTS useful in your research, please consider citing:

@inproceedings{yuan2026facts,
title={{FACTS}: Table Summarization via Offline Template Generation with Agentic Workflows},
author={Ye Yuan and Mohammad Amin Shabani and Siqi Liu},
booktitle={The 64th Annual Meeting of the Association for Computational Linguistics},
year={2026},
}