open2ch / README.md
p1atdev's picture
Upload README.md with huggingface_hub
8a48274
---
language:
- ja
license: apache-2.0
size_categories:
- 1M<n<10M
task_categories:
- text-generation
- text2text-generation
dataset_info:
- config_name: all-corpus
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
- name: board
dtype: string
splits:
- name: train
num_bytes: 1693355620
num_examples: 8134707
download_size: 868453263
dataset_size: 1693355620
- config_name: all-corpus-cleaned
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
- name: board
dtype: string
splits:
- name: train
num_bytes: 1199092499
num_examples: 6192730
download_size: 615570076
dataset_size: 1199092499
- config_name: livejupiter
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 1101433134
num_examples: 5943594
download_size: 592924274
dataset_size: 1101433134
- config_name: livejupiter-cleaned
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 807499499
num_examples: 4650253
download_size: 437414714
dataset_size: 807499499
- config_name: news4vip
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 420403926
num_examples: 1973817
download_size: 240974172
dataset_size: 420403926
- config_name: news4vip-cleaned
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 269941607
num_examples: 1402903
download_size: 156934128
dataset_size: 269941607
- config_name: newsplus
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 56071294
num_examples: 217296
download_size: 32368053
dataset_size: 56071294
- config_name: newsplus-cleaned
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
splits:
- name: train
num_bytes: 33387874
num_examples: 139574
download_size: 19556120
dataset_size: 33387874
- config_name: ranking
features:
- name: dialogue
sequence:
- name: speaker
dtype: int8
- name: content
dtype: string
- name: next
struct:
- name: speaker
dtype: int8
- name: content
dtype: string
- name: random
sequence: string
splits:
- name: train
num_bytes: 1605628
num_examples: 2000
- name: test
num_bytes: 1604356
num_examples: 1953
download_size: 2127033
dataset_size: 3209984
configs:
- config_name: all-corpus
data_files:
- split: train
path: all-corpus/train-*
- config_name: all-corpus-cleaned
data_files:
- split: train
path: all-corpus-cleaned/train-*
- config_name: livejupiter
data_files:
- split: train
path: livejupiter/train-*
- config_name: livejupiter-cleaned
data_files:
- split: train
path: livejupiter-cleaned/train-*
- config_name: news4vip
data_files:
- split: train
path: news4vip/train-*
- config_name: news4vip-cleaned
data_files:
- split: train
path: news4vip-cleaned/train-*
- config_name: newsplus
data_files:
- split: train
path: newsplus/train-*
- config_name: newsplus-cleaned
data_files:
- split: train
path: newsplus-cleaned/train-*
- config_name: ranking
data_files:
- split: train
path: ranking/train-*
- split: test
path: ranking/test-*
tags:
- not-for-all-audiences
---
# おーぷん2ちゃんねる対話コーパス
## Dataset Details
### Dataset Description
[おーぷん2ちゃんねる対話コーパス](https://github.com/1never/open2ch-dialogue-corpus) を Huggingface Datasets 向けに変換したものになります。
- **Curated by:** [More Information Needed]
- **Language:** Japanese
- **License:** Apache-2.0
### Dataset Sources
- **Repository:** https://github.com/1never/open2ch-dialogue-corpus
## Dataset Structure
- `all-corpus`: `livejupiter`, `news4vip`, `newsplus` サブセットを連結したもの
- `dialogue`: 対話データ (`list[dict]`)
- `speaker`: 話者番号。`1` または `2`
- `content`: 発言内容
- `board`: 連結元のサブセット名
- `livejupiter`: オリジナルのデータセットでの `livejupiter.tsv` から変換されたデータ。
- `dialogue`: 対話データ (`list[dict]`)
- `speaker`: 話者番号。`1` または `2`
- `content`: 発言内容
- `news4vip`: オリジナルのデータセットでの `news4vip.tsv` から変換されたデータ。
- 構造は同上
- `newsplus`: オリジナルのデータセットでの `newsplus.tsv` から変換されたデータ。
- 構造は同上
- `ranking`: 応答順位付けタスク用データ (オリジナルデータセットでの `ranking.zip`)
- `train``test` split があり、それぞれはオリジナルデータセットの `dev.tsv``test.tsv` に対応します。
- `dialogue`: 対話データ (`list[dict]`)
- `speaker`: 話者番号。`1` または `2`
- `content`: 発言内容
- `next`: 対話の次に続く正解の応答 (`dict`)
- `speaker`: 話者番号。`1` または `2`
- `content`: 発言内容
- `random`: ランダムに選ばれた応答 9 個 (`list[str]`)
また、`all-corpus`, `livejupiter`, `news4vip`, `newsplus` にはそれぞれ名前に `-cleaned` が付与されたバージョンがあり、これらのサブセットではオリジナルのデータセットで配布されていた NG ワードリストを利用してフィルタリングされたものです。
オリジナルのデータセットでは各発言内の改行は `__BR__` に置換されていますが、このデータセットではすべて `\n` に置き換えられています。
## Dataset Creation
### Source Data
(オリジナルのデータセットの説明より)
> おーぷん2ちゃんねるの「なんでも実況(ジュピター)」「ニュー速VIP」「ニュース速報+」の3つの掲示板をクロールして作成した対話コーパスです. おーぷん2ちゃんねる開設時から2019年7月20日までのデータを使用して作成しました.
#### Data Collection and Processing
[オリジナルのデータセット](https://github.com/1never/open2ch-dialogue-corpus) を参照。
#### Personal and Sensitive Information
`-cleaned` ではないサブセットでは、非常に不適切な表現が多いため注意が必要です。
## Usage
```py
from datasets import load_dataset
ds = load_dataset(
"p1atdev/open2ch",
name="all-corpus",
)
print(ds)
print(ds["train"][0])
# DatasetDict({
# train: Dataset({
# features: ['dialogue', 'board'],
# num_rows: 8134707
# })
# })
# {'dialogue': {'speaker': [1, 2], 'content': ['実況スレをたてる', 'おんj民の鑑']}, 'board': 'livejupiter'}
```