metadata

dataset_info:
  features:
    - name: question
      dtype: string
    - name: specific_question
      dtype: string
    - name: keyword
      dtype: string
    - name: answer
      dtype: string
    - name: last_usage
      dtype: timestamp[ns]
    - name: variants
      sequence: string
    - name: id
      dtype: int64
    - name: conversations
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
  splits:
    - name: train
      num_bytes: 13529402
      num_examples: 12363
  download_size: 7211893
  dataset_size: 13529402
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - question-answering
  - text2text-generation
  - text-generation
language:
  - cs
size_categories:
  - 10K<n<100K

Dataset Card for Dotazy of Institute of the Czech Language of the Czech Academy of Sciences (questions_ujc_cas_cs)

This is a dataset scraped from the webpage https://dotazy.ujc.cas.cz/ mantained by the Instute of the Czech Language of the Academy of Sciences of the Czech Republic.

Dataset Details

Dataset Description

This is a dataset scraped from the webpage https://dotazy.ujc.cas.cz/ mantained by the Instute of the Czech Language of the Academy of Sciences of the Czech Republic. It was scraped with their permission. It contains questions asked by the users of language consultancy of the Institute of the Czech Language along with answers and other metadata.

Curated by: AIC FEE CTU
Language(s) (NLP): Czech
License: CC-BY 4.0

Uses

This dataset is intended to be used in the process of instruction tuning of LLMs in Czech language. However, it can be utilized also for question-answering task or for language generation and other related tasks.

Dataset Structure

We provide only single split and encourage the users to modify the dataset as the need.

The dataset fields are:

question: more general question category
specific question: specific question that was asked
keyword: assigned keyword
answer: answer of the Instute
last_usage: date of the last usage
variants: variants of considered language problem
id: id number of the line

Dataset Creation

Curation Rationale

The creation was motivated by the scarcity of available high-quality instruction tuning dataset in Czech.

Source Data

The source is the language consultancy webpage: https://dotazy.ujc.cas.cz/

Data Collection and Processing

The dataset was scraped from the web and the strings were fixed using ftfy Python library.

Who are the source data producers?

The producers of source data are the users of the language consultancy and the employees of the Instute of the Czech Language - https://ujc.cas.cz/cs/

Personal and Sensitive Information

The dataset should not contain any personal or sensitive information according to the original description and the nature of the information inside of it.

Bias, Risks, and Limitations

The content of this dataset is limited to the Czech language and the topic is also just the Czech language.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset.

Citation [optional]

TBA