Augment_tableQA

This is the implementation for the paper: Augment before You Try: Knowledge-Enhanced Table Question Answering via Table Expansion.

Requirements

Environment

Install conda environment by running

conda env create -f environment.yml
conda activate augment

Usage

1) OpenAI key

Add your openai API keys in key.txt, one for each line.

2) Run scripts

The running scripts are provided in runscripts/. To run our method, please use run_augment_finqa.py, run_augment_tatqa.py, and run_augment_wikitq.py. The output will be stored in results/ and the performance will be printed. Note: We observe that there might be about 1% random performance variation even if we use greedy decoding. You might try to run the code again if you can't get the number reported in the paper.

References

If you find our work useful for your research, please consider citing our paper:

@misc{liu2024augment,
      title={Augment before You Try: Knowledge-Enhanced Table Question Answering via Table Expansion}, 
      author={Yujian Liu and Jiabao Ji and Tong Yu and Ryan Rossi and Sungchul Kim and Handong Zhao and Ritwik Sinha and Yang Zhang and Shiyu Chang},
      year={2024},
      eprint={2401.15555},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Our implementation is based on the following repos:

• https://github.com/xlang-ai/Binder
• https://github.com/wenhuchen/Program-of-Thoughts