README.md

---
license: mit
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: pdf
    dtype: pdf
  - name: html
    dtype: string
  - name: question
    dtype: string
  - name: answer
    dtype: string
  - name: title
    dtype: string
  splits:
  - name: train
    num_bytes: 5833510
    num_examples: 300
  download_size: 27811899
  dataset_size: 5833510
task_categories:
- table-question-answering
language:
- ru
size_categories:
- n<1K
---
# rosstat-table-qa

`rosstat-table-qa` is a Russian-language dataset for question answering over statistical tables from Rosstat (Federal State Statistics Service of Russia).

The dataset contains real-world statistical tables from official Rosstat Excel reports. Each worksheet was converted into a separate table sample represented as both PDF and HTML.

Each sample includes:

- the original PDF table (`pdf`),
- an HTML representation of the table (`html`),
- the table title (`title`),
- a natural language question (`question`),
- and the corresponding answer (`answer`).


## Dataset Structure

Each sample has the following structure:

```python
{
    "pdf": PDF,
    "html": str,
    "title": str,
    "question": str,
    "answer": str
}
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("gu1show/rosstat-table-qa")

print(dataset["train"][0])
```
Output:
```
{'pdf': <pdfplumber.pdf.PDF at 0x15115ff3e30>,
 'html': '<table>...</table>',
 'question': 'Какое количество гектаров лесов обработали биологическими препаратами для уничтожения вредных организмов наземным способом в 2021 году?',
 'answer': '60,4 тысяч гектаров',
 'title': 'Защита лесов в Российской Федерации (тысяч гектаров)'}
```