--- dataset_info: - config_name: cognitive_error features: - name: page dtype: string - name: title dtype: string - name: pre_str dtype: string - name: post_str dtype: string - name: pre_bart_likelihood dtype: float64 - name: post_bart_likelihood dtype: float64 - name: category dtype: string - name: text_head dtype: string - name: text_tail dtype: string - name: typo_type dtype: string - name: uuid dtype: string splits: - name: train num_bytes: 23497801 num_examples: 65052 - name: test num_bytes: 7229701 num_examples: 20000 download_size: 19894192 dataset_size: 30727502 - config_name: cognitive_error_v2 features: - name: page dtype: string - name: title dtype: string - name: pre_str dtype: string - name: post_str dtype: string - name: pre_bart_likelihood dtype: float64 - name: post_bart_likelihood dtype: float64 - name: category dtype: string - name: text_head dtype: string - name: text_tail dtype: string - name: typo_type dtype: string - name: uuid dtype: string splits: - name: train num_bytes: 20368950.96830228 num_examples: 56390 - name: test num_bytes: 6275741.95305 num_examples: 17361 download_size: 16839387 dataset_size: 26644692.92135228 - config_name: default features: - name: page dtype: string - name: title dtype: string - name: pre_str dtype: string - name: post_str dtype: string - name: pre_bart_likelihood dtype: float64 - name: post_bart_likelihood dtype: float64 - name: category dtype: string - name: text_head dtype: string - name: text_tail dtype: string splits: - name: train num_bytes: 79765386 num_examples: 291954 - name: test num_bytes: 584213 num_examples: 2100 download_size: 46334306 dataset_size: 80349599 - config_name: labeled features: - name: text_head dtype: string - name: pre_str dtype: string - name: post_str dtype: string - name: text_tail dtype: string - name: error_type dtype: string - name: category dtype: string - name: page dtype: int64 - name: title dtype: string - name: pre_bart_likelihood dtype: float64 - name: post_bart_likelihood dtype: float64 - name: diff_bart_likelihood dtype: float64 splits: - name: train num_bytes: 385103 num_examples: 1300 download_size: 248376 dataset_size: 385103 - config_name: labeled_llm features: - name: page dtype: string - name: title dtype: string - name: pre_str dtype: string - name: post_str dtype: string - name: pre_bart_likelihood dtype: float64 - name: post_bart_likelihood dtype: float64 - name: category dtype: string - name: text_head dtype: string - name: text_tail dtype: string - name: typo_type dtype: string splits: - name: train num_bytes: 85027947 num_examples: 291954 download_size: 46090653 dataset_size: 85027947 configs: - config_name: cognitive_error data_files: - split: train path: cognitive_error/train-* - split: test path: cognitive_error/test-* - config_name: cognitive_error_v2 data_files: - split: train path: cognitive_error_v2/train-* - split: test path: cognitive_error_v2/test-* - config_name: default data_files: - split: train path: data/train-* - split: test path: data/test-* - config_name: labeled data_files: - split: train path: labeled/train-* - config_name: labeled_llm data_files: - split: train path: labeled_llm/train-* license: cc-by-sa-3.0 language: - ja --- # 日本語ウィキペディア入力誤りデータセット (漢字誤変換抽出版) ## 概要 このデータセットは,京都大学 言語メディア研究室によって公開されているデータセットをHuggingFaceで使用できるよう変換したものです. [日本語Wikipedia入力誤りデータセット (v2) - 京都大学 言語メディア研究室](https://nlp.ist.i.kyoto-u.ac.jp/?%E6%97%A5%E6%9C%AC%E8%AA%9EWikipedia%E5%85%A5%E5%8A%9B%E8%AA%A4%E3%82%8A%E3%83%87%E3%83%BC%E3%82%BF%E3%82%BB%E3%83%83%E3%83%88) このデータセットでは,[JunSotohigashi/JapaneseWikipediaTypoDataset](https://huggingface.co/datasets/JunSotohigashi/JapaneseWikipediaTypoDataset)から,漢字誤変換のみを抽出しています. 誤りの種類が`kanji-conversion_a`と`kanji-conversion_b`の事例のみをフィルタリングし,誤り部分の前と後ろに文を分割しました. これにより,`pre_text`は`text_head + pre_str + text_tail`と等価になります. `post_text`については,diffsに複数のデータが存在する場合にはそれをすべて適用したあとの文字列であるため,`post_text`と`text_head + post_str + text_tail`が同じになるとは限りません. 生成に使用したソースコードは[ここ](https://github.com/JunSotohigashi/misusing-corpus-jp/tree/v2.2)から参照できます. ## サブセット ### default オリジナルから,`kanji-conversion_a`と`kanji-conversion_b`のみ取り出し ### labeled trainデータの一部に対して,人手により誤りタイプのタグをつけました. - trainデータをseed=42でシャッフル - 先頭から1300件抽出 - error_type列に,勘違いに起因するタイポ(CognitiveError)と,単純なキーの押し間違い(KeystrokeError)の分類結果を保存 - 分類時に与えるプロンプトの設計に使用 ### labeled_llm labeledを使用して設計したプロンプトを使用し,defaultのtrainデータ全件に対して誤り分類ラベルを付与 ### cognitive_error labeled_llmデータのうち,CognitiveErrorに分類され,text_headが20文字より長い事例のみ抽出 - seed=100でシャッフル - testデータとして20000件をsplit ## ライセンス 配布元に従い,CC-BY-SA 3.0とします.