readme.md

The dataset and processing code for IJCNLP-AACL 2023 paper “Enhancing Open-Domain Table Question Answering via Syntax- and Structure-aware Dense Retrieval”.

Prepare dataset

1. Download the original dataset

Please download the original WikiSQL and NQ-Tables . Then decompress and move them to data/wikisql and data/nqtables folders, respectively.

data
└───nqtable
│   └───interactions
│       │   ...
|   └───tables
|       |   ...
│   
└───wikisql
    │   dev.db
    │   dev.jsonl
    |   dev.tables.jsonl
    |   ...

2. preprocess the original dataset

cd ODTQA
bash script/run_wikisql_preprocess.sh
bash script/run_nqtable_preprocess.sh

The processed datasets for retrieval will be saved in nqtable/processed folder and wikisql/processed folder.

Mention: you can also download the processed dataset directly from here.