metadata

language:
  - en
  - es
pretty_name: ' 💾🏋️💾 DataBench 💾🏋️💾'
tags:
  - table-question-answering
  - table
  - qa
license: mit
task_categories:
  - table-question-answering
  - question-answering
default: qa
configs:
  - config_name: qa
    data_files:
      - ./066_40db_Igualdad/qa.parquet
      - ./067_40dB_Dormir/qa.parquet
      - ./068_CIS_Enero_Marzo_2023/qa.parquet
      - ./069_CEA_Barometro_Andaluz_Septiembre_2023/qa.parquet
      - ./070_CIS_2023_Salud_Bienestar/qa.parquet
      - ./071_CIS_Politica_Fiscal_Julio_2023/qa.parquet
      - ./072_CIS_Relaciones_Afectivas_Pospandemia_III/qa.parquet
      - ./073_CIS_Barometro_Diciembre_2022/qa.parquet
      - ./074_40dB_Percepcion_Amor/qa.parquet
      - >-
        ./075_CIS_Salud_Mental_Pandemia_2021/qa.parquet     -
        ./066_Influencers/qa.parquet
      - ./066_Clustering/qa.parquet
      - ./066_RFM/qa.parquet

💾🏋️💾 DataBench 💾🏋️💾

This repository contains the original 65 datasets used for the paper Towards Quality Benchmarking in Question Answering over Tabular Data in Spanish which appeared in SEPLN 2024.

It is a spin-off of the original suite in English, which you can find here.

Spa-DataBench brings together ten tabular datasets from Spain’s major survey agencies: CIS, CEA, CRS, and 40dB. These datasets, initially available publicly, have been unified and enriched with our typing system (see Section 3.2) to simplify processing. Designed as an evaluation benchmark, each dataset is paired with twenty uniquely crafted questions along with their corresponding gold answers, totaling 200 Q&A pairs. This tuple-based structure (dataset, questions, answers) makes it easy to incorporate new datasets.

Usage

from datasets import load_dataset

# Load all QA pairs
all_qa = load_dataset("SINAI/databenchSPA")

You can use any of the individual integrated libraries to load the actual data where the answer is to be retrieved.

For example, using pandas in Python:

import pandas as pd

# "001_Forbes", the id of the dataset
ds_id = all_qa['dataset'][0] 

# full dataset
df = pd.read_parquet(f"hf://datasets/SINAI/databenchSPA/{ds_id}/all.parquet")

# sample dataset
df = pd.read_parquet(f"hf://datasets/SINAI/databenchSPA/{ds_id}/sample.parquet")

#	Name	Rows	Columns	#QA	Source (Reference)
1	Encuesta de Igualidad	2000	105	20	40dB
2	Calidad del Sueño	2000	80	20	40dB
3	Fusión Barómetros	7430	161	20	CIS
4	Barómetro Andaluz	5349	85	20	CEA
5	Juventud	1510	236	20	CRS
6	Política Fiscal	3011	198	20	CIS
7	Relaciones	2491	186	20	CIS
8	Barómetro Mensual	2444	185	20	CIS
9	Percepción del Amor	2000	150	20	40dB
10	Salud Mental	3083	354	20	CIS
	Total	31318	1741	200



## 🏗️ Folder structure
Each folder represents one dataset. You will find the following files within:

* all.parquet: the processed data, with each column tagged with our typing system, in [parquet](https://arrow.apache.org/docs/python/parquet.html).
* qa.parquet: contains the human-made set of questions, tagged by type, for the dataset (sample_answer indicates the answers for DataBench lite)
* info.yml: additional information about the dataset

## 🗂️ Column typing system
In an effort to map the stage for later analysis, we have categorized the columns by type. This information allows us to segment different kinds of data so that we can subsequently analyze the model's behavior on each column type separately. All parquet files have been casted to their smallest viable data type using the open source [Lector](https://github.com/graphext/lector) reader.

What this means is that in the data types we have more granular information that allows us to know if the column contains NaNs or not (following panda’s convention of Int vs int), as well as whether small numerical values contain negatives (Uint vs int) and their range. We also have dates with potential timezone information (although for now they’re all UTC), as well as information about categories’ cardinality coming from the arrow types.

In the table below you can see all the data types assigned to each column, as well as the number of columns for each type. The most common data types are numbers and categories with 1336 columns of the total of 1615 included in DataBench. These are followed by some other more rare types as urls, booleans, dates or lists of elements.

| Type           | Columns | Example                 |
| -------------- | ------- | ----------------------- |
| number         | 269   | 1                     |
| category       | 1464  | banana                |
| date           | 2     | 1979-01-01            |
| text           | 1     | A blue rabbit went to... |
| list[number]   | 1     | [10,11,12]            |
| list[category] | 4     | [banana, pineapple]   |

## 🔗 Reference

You can download the paper [here](http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6617).

If you use this resource, please use the following reference:

@article{DBLP:journals/pdln/GrijalbaLCC24, author={Jorge Osés Grijalba and Luis Alfonso Ureña López and José Camacho-Collados and Eugenio Martínez Cámara}, title={Towards Quality Benchmarking in Question Answering over Tabular Data in Spanish}, year={2024}, cdate={1704067200000}, journal={Proces. del Leng. Natural}, volume={73}, pages={283-296}, url={http://journal.sepln.org/sepln/ojs/ojs/index.php/pln/article/view/6617} } ```