taiga_stripped_rest / README.md
cointegrated's picture
Update README.md
f1c5815
---
dataset_info:
features:
- name: text
dtype: string
- name: file
dtype: string
splits:
- name: Arzamas
num_bytes: 4725465
num_examples: 311
- name: Interfax
num_bytes: 82478694
num_examples: 46000
- name: Lenta
num_bytes: 99984639
num_examples: 36000
- name: Magazines
num_bytes: 2295653294
num_examples: 39000
- name: NPlus1
num_bytes: 23506941
num_examples: 7000
- name: KP
num_bytes: 65444392
num_examples: 45000
- name: Fontanka
num_bytes: 840679591
num_examples: 342683
- name: Subtitles
num_bytes: 311508573
num_examples: 7903
- name: social
num_bytes: 600396164
num_examples: 804356
download_size: 2180717682
dataset_size: 4324377753
license: cc-by-sa-3.0
language:
- ru
tags:
- taiga
- tayga
size_categories:
- 1M<n<10M
task_categories:
- text-generation
- fill-mask
---
# Dataset Card for "taiga_stripped_rest"
This is a subset of the Taiga corpus (https://tatianashavrina.github.io/taiga_site), derived from the all the sources, except
[stihi](https://huggingface.co/datasets/cointegrated/taiga_stripped_stihi) and [proza](https://huggingface.co/datasets/cointegrated/taiga_stripped_proza):
`Arzamas`, `Interfax`, `Lenta`, `Magazines`, `NPlus1`, `KP`, `Fontanka`, `Subtitles` and `social`.
The dataset consists of plain texts, without morphological and syntactic annotation or metainformation.
For the `Subtitles` subset, we dropped all non-Russian texts.
For the `social` subset, we split the texts into indidividual database items,
or (for LiveJournal) into "posts" (defined as lines with 1000+ characters) and subsequent "comments".
For more details and analysis, and for the texts with annotation or metadata, please refer to website of the corpus.
Other subsets of Taiga: [proza](https://huggingface.co/datasets/cointegrated/taiga_stripped_proza) (fiction)
and [stihi](https://huggingface.co/datasets/cointegrated/taiga_stripped_stihi) (poetry).
License: [CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/).