slovak-nlp/e-testy-sk

e-testy.sk Dataset

Dataset Summary

A multimodal question-answering dataset of Slovak secondary-school test questions extracted from e-testy.sk, a free Slovak educational testing portal. It contains 1,834 questions from 246 tests across 9 school subjects, with single-choice, multiple-choice, and text-input question types. A small subset of questions (27) includes an associated image (anatomy diagrams, hardware schematics, network topologies, etc.), enabling multimodal evaluation.

• Language: Slovak (sk)
• Total questions: 1,834 (single split: train)
• Tests: 246
• Question types: single_choice (960), text_input (448), multiple_choice (426)
• Questions with images: 27
• Questions with extracted correct answers: 1,784 (97.3%)

Subjects

Subject (subject)	Tests	Questions	Images
slovensky-jazyk (Slovak language and literature)	93	799	0
informatika (Informatics)	42	262	11
biologia (Biology)	34	196	16
chemia (Chemistry)	34	156	0
obcianska-nauka (Civics)	12	111	0
geografia (Geography)	10	100	0
dejepis (History)	10	100	0
fyzika (Physics)	7	70	0
matematika (Mathematics)	4	40	0
Total	246	1,834	27

Supported Tasks

• Multiple-choice / single-choice question answering — benchmarking Slovak LLMs on secondary-school curriculum knowledge.
• Open-ended question answering — the text_input subset requires free-form answers.
• Visual question answering (VQA) — the image-grounded subset (has_image = True) for Slovak multimodal models.

Dataset Structure

Data Instances

{
    "subject": "biologia",
    "test_path": "/biologia/bakterie",
    "test_name": "bakterie",
    "question_number": 1,
    "question_type": "single_choice",
    "question": "Vyber správnu možnosť, ktorá uvádza približnú veľkosť baktérií",
    "correct_answers": ["0,2 - 2μm"],
    "options": '[{"text": "0,2 - 2μm", "correct": true}, {"text": "0,01 - 1μm", "correct": false}, ...]',
    "image": None,
    "image_url": None,
    "has_image": False,
    "has_correct_answer": True
}

Data Fields

Field	Type	Description
subject	string	Subject area (e.g., biologia, informatika, slovensky-jazyk).
test_path	string	URL path of the source test on e-testy.sk.
test_name	string	Short test identifier.
question_number	int64	Question number within the test.
question_type	string	One of single_choice, multiple_choice, text_input.
question	string	Question text in Slovak.
correct_answers	list[string]	Correct answer string(s). Multiple entries for multiple_choice questions.
options	string (JSON)	JSON-encoded list of {"text": "...", "correct": bool} for choice questions; null for text_input.
image	image (nullable)	Associated diagram/image, or None.
image_url	string	Original image URL on e-testy.sk (if any).
has_image	bool	Whether the question has an associated image.
has_correct_answer	bool	Whether correct answers were successfully extracted.

Data Splits

Split	Examples
train	1,834

The single split is intended to be used as an evaluation set (the name train only reflects the default split name).

How to Use

import json
from datasets import load_dataset

ds = load_dataset("slovak-nlp/e-testy-sk")

# Filter by subject
bio = ds["train"].filter(lambda x: x["subject"] == "biologia")

# Only questions with extracted correct answers
answered = ds["train"].filter(lambda x: x["has_correct_answer"])

# Multimodal questions
with_images = ds["train"].filter(lambda x: x["has_image"])

# Parse the JSON-encoded options
example = ds["train"][0]
options = json.loads(example["options"]) if example["options"] else None

Dataset Creation

All data was extracted from e-testy.sk, a free Slovak educational testing portal for secondary-school (stredná škola / gymnázium) students and teachers. For each question, the question text, answer options with correctness flags, correct answers, and any associated image were collected and parsed into a structured format.

Considerations for Using the Data

• 50 questions (2.7%) have empty correct_answers (has_correct_answer = False) — these are image-labeling exercises where answers map to numbered positions on a diagram. Filter them out for standard QA evaluation.
• The options field is a JSON-encoded string, not a native list — parse it with json.loads before use.
• For multiple_choice questions, all entries in correct_answers must be selected for a response to count as correct.
• Image-grounded questions (has_image = True) require a multimodal model; text-only models should be evaluated on the has_image = False subset.

License

The dataset is derived from publicly available educational content on e-testy.sk. It is distributed for research and evaluation purposes; please respect the original source's terms of use.

Citation

If you use this dataset, please cite it as:

@misc{e_testy_sk,
  title        = {e-testy.sk Dataset},
  author       = {{Slovak NLP}},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/slovak-nlp/e-testy-sk}}
}