rokayon/AgriQA_Multilingual_Dataset

AgriQA Multilingual Dataset

A multilingual agricultural question-answering dataset derived from the KisanVaani corpus, covering four languages: Bengali (bn), English (en), Spanish (es), and Chinese (zh).

Dataset Statistics

Split	Rows
Train	72,368
Test	18,092
Total	90,460

Language Distribution (per split — stratified)

Language	Train	Test
Bengali (bn)	18,092	4,523
English (en)	18,092	4,523
Spanish (es)	18,092	4,523
Chinese (zh)	18,092	4,523

Dataset Structure

from datasets import load_dataset

ds = load_dataset("rokayon/AgriQA_Multilingual_Dataset")

Fields

Column	Type	Description
id	string	Unique identifier for each example
question	string	Agricultural question in the target language
answer	string	Answer to the question in the target language
lang	string	ISO 639-1 language code (bn, en, es, zh)
text	string	Combined question + answer text

Splits

The dataset is split 80 % train / 20 % test using a stratified split so that each language contributes equally proportional rows to both splits.

License

CC BY 4.0

Citation

If you use this dataset, please cite the original KisanVaani work and this HuggingFace repository:

@dataset{rokayon_agriqamultilingual_2025,
  author    = {Rokayon},
  title     = {AgriQA Multilingual Dataset},
  year      = {2025},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/rokayon/AgriQA_Multilingual_Dataset}
}