hyukkyukang / table-to-text

Table-to-Text Summarization Model

Table-to-text

This is a Python program that creates a summarizing text from a table.

Setting environment

1. Run docker container using docker compose

docker compose up -d

2. Download totto dataset and evaluation scripts

sh script/download_totto.sh
pip install -r evaluation/language_repo/language/totto/eval_requirements.txt
cd evaluation/language_repo/ && python setup.py install

Training

Run python srcipts in src directory.

python src/train.py

Inference

To run a Table-to-text inference, you must train a model first. Run python srcipts in src directory. Note that the guiding text is optional.

python src/infer.py --input [string of table] --saved_path [location of saved model] --guidance [guiding text]

Evaluation

For the evaluation, we conducted a user study following the Provenance for natural language queries (VLDB'17) paper. The pdf in the repository contains the questions (query, result table, and natural language query) for the conducted user study.