Datasets:
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
fever/feverous(複製)
データの内容
Wikipedia の文と表のセルを根拠に、主張の正誤を判定するファクトチェックのデータ。87,026 件の主張に、根拠が支持するか、反論するか、判断に十分な情報がないかのラベルが付いている。1 件の例は claim、label、evidence などからなり、注釈時の検索キーワードやページのクリック、時刻の記録も含む。
データの概要
| 項目 | 値 |
|---|---|
| 行数・列数 | 不明(データセットビューアで読めない形式。ファイルの内訳だけを示す) |
| 容量 | 20.49 KB(ファイル 3 本) |
| ファイルの形式 | py 1 |
| 言語 | en |
| 公開 | public |
全データセットの一覧は HYUKI5/dataset-catalog にある。
複製について
fever/feverous をリビジョン 96a6c960623e1b4ad83b38f5e345c9c5632857f7 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。
出典
- 元のデータセット: https://huggingface.co/datasets/fever/feverous
- 複製したリビジョン:
96a6c960623e1b4ad83b38f5e345c9c5632857f7(2022-10-25 05:50:36+00:00) - 元のライセンス表記:
['cc-by-sa-3.0']
論文
アノテーションの方法
収集の根拠
初期データの収集と正規化
ソース言語の作成者は誰か?
アノテーションの過程
注釈者は誰か?
微調整に使うときの注意
- 調査メモの台帳での区分: B(継承条件・gated・モデル由来の条件・上流の規約などが付く)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
- Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
- 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。
調査メモの記述
| メモ | 節 | ライセンス(記載) | 確認 | 注記 |
|---|---|---|---|---|
reasoning_argument_numeric.md |
Q5. 表を根拠にした主張の正誤(事実検証)のラベルが付いたデータは何があるか (L130) | cc-by-sa-3.0(HF)。arXiv ページの CC BY-SA 4.0 は論文本体の表記 | 確認済 | 複数/混在 / 原文注記: script(feverous.py)。第三者の整形版は Dzeniks/feverous_3way(mit、DL 192)などがあるが内容は未確認。既出の FEVER は general_english_nlu.md |
元のカード(日本語訳)
以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。
FEVEROUS のデータセットカード
目次
- データセットの説明
- データセットの概要
- 対応タスクとリーダーボード
- 言語
- データセットの構造
- データのインスタンス
- データのフィールド
- データの分割
- データセットの作成
- 収集の根拠
- ソースデータ
- アノテーション
- 個人情報と機微な情報
- データ利用上の留意点
- データセットの社会的影響
- バイアスに関する議論
- その他の既知の制約
- 追加情報
- データセットのキュレーター
- ライセンス情報
- 引用情報
- 貢献者
データセットの説明
- ホームページ: https://fever.ai/dataset/feverous.html
- リポジトリ: More Information Needed
- 論文: FEVEROUS: Fact Extraction and VERification Over Unstructured and Structured information
- 連絡先: More Information Needed
データセットの概要
ウェブには、ほぼあらゆる話題について情報を提供する数十億のページがあり、ほぼあらゆる問いに答える事実を集められるはずである。しかし、その情報のうち構造化された情報源(Wikidata、Freebase など)に含まれるのは、ごく一部にすぎない。そのため、自由形式のテキストを構造化された知識へ変換する能力によって、私たちは制約を受けている。さらに、近年の研究や報道の多くが焦点を当てている別の問題もある。信頼できない情報源から出てくる誤った情報である。
FEVER ワークショップは、検証可能な知識抽出に関する研究の場であり、この方向の進展を促すためのものである。
FEVEROUS(Fact Extraction and VERification Over Unstructured and Structured information)は、87,026 件の検証済みの主張からなる事実検証データセットである。各主張には、Wikipedia の文と表のセルの形で根拠が付されており、根拠がその主張を支持するか、反論するか、判断に十分な情報がないかを示すラベルも付いている。このデータセットには、注釈者の操作(検索キーワード、ページのクリック、時刻の記録)や、各主張が示す課題の種類といった注釈のメタデータも含まれる。
対応タスクとリーダーボード
このタスクは、テキストの主張を、テキストの情報源と照らして検証することである。
テキスト含意(textual entailment、TE)・自然言語推論と比べると、重要な違いは次の点にある。これらのタスクでは、主張を検証する対象の文章が与えられており、近年は通常それが 1 文からなる。一方、検証システムでは、根拠を形成するためにその文章を大量の文書から検索して得る。
言語
データセットは英語(en)である。
データセットの構造
データのインスタンス
- ダウンロードしたデータセットファイルのサイズ: 187.82 MB
- 生成されたデータセットのサイズ: 123.25 MB
- ディスク使用量の合計: 311.07 MB
'wikipedia_pages' の例は次のとおり。
{'id': 24435,
'label': 1,
'claim': 'Michael Folivi competed with ten teams from 2016 to 2021, appearing in 54 games and making seven goals in total.',
'evidence': [{'content': ['Michael Folivi_cell_1_2_0',
'Michael Folivi_cell_1_7_0',
'Michael Folivi_cell_1_8_0',
'Michael Folivi_cell_1_9_0',
'Michael Folivi_cell_1_12_0'],
'context': [['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0']]},
{'content': ['Michael Folivi_cell_0_13_1',
'Michael Folivi_cell_0_14_1',
'Michael Folivi_cell_0_15_1',
'Michael Folivi_cell_0_16_1',
'Michael Folivi_cell_0_18_1'],
'context': [['Michael Folivi_title',
'Michael Folivi_header_cell_0_13_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_14_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_15_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_16_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_18_0',
'Michael Folivi_header_cell_0_11_0']]}],
'annotator_operations': [{'operation': 'start',
'value': 'start',
'time': 0.0},
{'operation': 'Now on', 'value': '?search=', 'time': 0.78},
{'operation': 'search', 'value': 'Michael Folivi', 'time': 78.101},
{'operation': 'Now on', 'value': 'Michael Folivi', 'time': 78.822},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_2_0',
'time': 96.202},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_7_0',
'time': 96.9},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_8_0',
'time': 97.429},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_9_0',
'time': 97.994},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_12_0',
'time': 99.02},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_13_1',
'time': 106.108},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_14_1',
'time': 106.702},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_15_1',
'time': 107.423},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_16_1',
'time': 108.186},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_17_1',
'time': 108.788},
{'operation': 'Highlighting',
'value': 'Michael Folivi_header_cell_0_17_0',
'time': 108.8},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_18_1',
'time': 109.469},
{'operation': 'Highlighting deleted',
'value': 'Michael Folivi_cell_0_17_1',
'time': 124.28},
{'operation': 'Highlighting deleted',
'value': 'Michael Folivi_header_cell_0_17_0',
'time': 124.293},
{'operation': 'finish', 'value': 'finish', 'time': 141.351}],
'expected_challenge': '',
'challenge': 'Numerical Reasoning'}
データのフィールド
データのフィールドは、すべての split で共通である。
id(int): サンプルの ID。label(ClassLabel): 主張に付けられた注釈ラベル。{"SUPPORTS", "REFUTES", "NOT ENOUGH INFO"} のいずれかになる。claim(str): 主張の文。evidence(list of dict): 根拠の集合(最大 3 つ)。各集合は、次の 2 つのフィールドを持つ辞書からなる:content(list of str): 主張の根拠となる要素 ID のリスト。各要素 ID は次の形式を取る"[PAGE ID]_[EVIDENCE TYPE]_[NUMBER ID]"。[EVIDENCE TYPE]はsentence、cell、header_cell、table_caption、itemのいずれか。context(list of list of str):contentの各要素 ID について、その ID に自動的に結び付けられ、文脈として使われる Wikipedia の要素のリスト。記事の題名、関連する節(その要素が位置する節と下位の節)を含む。セルの場合は、最も近い行見出しと列見出しも含む(見出しが連続していれば複数の行・列見出し)。
annotator_operations(list of dict): 注釈者が、主張に対して根拠を探し、判定に至るまでに使った操作のリスト。リストの各要素は次のフィールドを持つ辞書である:operation(str): 操作の名前。次のいずれか:start、finish: 注釈が始まった・終わったことを表す。value は操作の名前である。search: 注釈者が Wikipedia の検索機能を使ったことを表す。value は入力された検索語、または自動候補から選ばれた語である。候補を選ばず詳細検索に進んだ場合は、語の前に "contains..." が付く。hyperlink: 注釈者がページ内のハイパーリンクをクリックしたことを表す。value はそのリンクのアンカーテキストである。Now on: 検索またはハイパーリンクのクリックの後に、注釈者がたどり着いたページを表す。value は PAGE ID である。Page search: 注釈者がページ内を検索したことを表す。value は検索語である。page-search-reset: 注釈者が検索欄を消去したことを表す。value は操作の名前である。Highlighting、Highlighting deleted: 注釈者がページ上の要素を選択・選択解除したことを表す。value はELEMENT IDである。back-button-clicked: 注釈者が戻るボタンを押したことを表す。value は操作の名前である。
value(str): 操作に伴う値。time(float): 注釈の開始からの経過時間(秒)。
expected_challenge(str): 主張の作成者が選んだ、主張を検証する際に直面すると想定される課題。次のいずれか:Numerical Reasoning、Multi-hop Reasoning、Entity Disambiguation、Combining Tables and Text、Search terms not in claim、Other。challenge(str): 主張を検証する際の主な課題。次のいずれか:Numerical Reasoning、Multi-hop Reasoning、Entity Disambiguation、Combining Tables and Text、Search terms not in claim、Other。
データの分割
| train | validation | test | |
|---|---|---|---|
| サンプル数 | 71291 | 7890 | 7845 |
データセットの作成
収集の根拠
ソースデータ
初期データの収集と正規化
ソース言語の作成者は誰か?
アノテーション
アノテーションの過程
注釈者は誰か?
個人情報と機微な情報
データ利用上の留意点
データセットの社会的影響
バイアスに関する議論
その他の既知の制約
追加情報
データセットのキュレーター
ライセンス情報
These data annotations incorporate material from Wikipedia, which is licensed pursuant to the Wikipedia Copyright Policy. These annotations are made available under the license terms described on the applicable Wikipedia article pages, or, where Wikipedia license terms are unavailable, under the Creative Commons Attribution-ShareAlike License (version 3.0), available at http://creativecommons.org/licenses/by-sa/3.0/ (collectively, the “License Termsâ€). You may not use these files except in compliance with the applicable License Terms.
引用情報
このデータセットを使う場合は、次を引用すること。
@inproceedings{Aly21Feverous,
author = {Aly, Rami and Guo, Zhijiang and Schlichtkrull, Michael Sejr and Thorne, James and Vlachos, Andreas and Christodoulopoulos, Christos and Cocarascu, Oana and Mittal, Arpit},
title = {{FEVEROUS}: Fact Extraction and {VERification} Over Unstructured and Structured information},
eprint={2106.05707},
archivePrefix={arXiv},
primaryClass={cs.CL},
year = {2021}
}
貢献者
このデータセットを追加してくれた @albertvillanova に感謝する。
原文(英語)
Dataset Card for FEVEROUS
Dataset Summary
With billions of individual pages on the web providing information on almost every conceivable topic, we should have the ability to collect facts that answer almost every conceivable question. However, only a small fraction of this information is contained in structured sources (Wikidata, Freebase, etc.) – we are therefore limited by our ability to transform free-form text to structured knowledge. There is, however, another problem that has become the focus of a lot of recent research and media coverage: false information coming from unreliable sources.
The FEVER workshops are a venue for work in verifiable knowledge extraction and to stimulate progress in this direction.
FEVEROUS (Fact Extraction and VERification Over Unstructured and Structured information) is a fact verification dataset which consists of 87,026 verified claims. Each claim is annotated with evidence in the form of sentences and/or cells from tables in Wikipedia, as well as a label indicating whether this evidence supports, refutes, or does not provide enough information to reach a verdict. The dataset also contains annotation metadata such as annotator actions (query keywords, clicks on page, time signatures), and the type of challenge each claim poses.
Supported Tasks and Leaderboards
The task is verification of textual claims against textual sources.
When compared to textual entailment (TE)/natural language inference, the key difference is that in these tasks the passage to verify each claim is given, and in recent years it typically consists a single sentence, while in verification systems it is retrieved from a large set of documents in order to form the evidence.
Languages
The dataset is in English (en).
Dataset Structure
Data Instances
- Size of downloaded dataset files: 187.82 MB
- Size of the generated dataset: 123.25 MB
- Total amount of disk used: 311.07 MB
An example of 'wikipedia_pages' looks as follows:
{'id': 24435,
'label': 1,
'claim': 'Michael Folivi competed with ten teams from 2016 to 2021, appearing in 54 games and making seven goals in total.',
'evidence': [{'content': ['Michael Folivi_cell_1_2_0',
'Michael Folivi_cell_1_7_0',
'Michael Folivi_cell_1_8_0',
'Michael Folivi_cell_1_9_0',
'Michael Folivi_cell_1_12_0'],
'context': [['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0'],
['Michael Folivi_title',
'Michael Folivi_section_4',
'Michael Folivi_header_cell_1_0_0']]},
{'content': ['Michael Folivi_cell_0_13_1',
'Michael Folivi_cell_0_14_1',
'Michael Folivi_cell_0_15_1',
'Michael Folivi_cell_0_16_1',
'Michael Folivi_cell_0_18_1'],
'context': [['Michael Folivi_title',
'Michael Folivi_header_cell_0_13_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_14_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_15_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_16_0',
'Michael Folivi_header_cell_0_11_0'],
['Michael Folivi_title',
'Michael Folivi_header_cell_0_18_0',
'Michael Folivi_header_cell_0_11_0']]}],
'annotator_operations': [{'operation': 'start',
'value': 'start',
'time': 0.0},
{'operation': 'Now on', 'value': '?search=', 'time': 0.78},
{'operation': 'search', 'value': 'Michael Folivi', 'time': 78.101},
{'operation': 'Now on', 'value': 'Michael Folivi', 'time': 78.822},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_2_0',
'time': 96.202},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_7_0',
'time': 96.9},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_8_0',
'time': 97.429},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_9_0',
'time': 97.994},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_1_12_0',
'time': 99.02},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_13_1',
'time': 106.108},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_14_1',
'time': 106.702},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_15_1',
'time': 107.423},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_16_1',
'time': 108.186},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_17_1',
'time': 108.788},
{'operation': 'Highlighting',
'value': 'Michael Folivi_header_cell_0_17_0',
'time': 108.8},
{'operation': 'Highlighting',
'value': 'Michael Folivi_cell_0_18_1',
'time': 109.469},
{'operation': 'Highlighting deleted',
'value': 'Michael Folivi_cell_0_17_1',
'time': 124.28},
{'operation': 'Highlighting deleted',
'value': 'Michael Folivi_header_cell_0_17_0',
'time': 124.293},
{'operation': 'finish', 'value': 'finish', 'time': 141.351}],
'expected_challenge': '',
'challenge': 'Numerical Reasoning'}
Data Fields
The data fields are the same among all splits.
id(int): ID of the sample.label(ClassLabel): Annotated label for the claim. Can be one of {"SUPPORTS", "REFUTES", "NOT ENOUGH INFO"}.claim(str): Text of the claim.evidence(list of dict): Evidence sets (at maximum three). Each set consists of dictionaries with two fields:content(list of str): List of element IDs serving as the evidence for the claim. Each element ID is in the format"[PAGE ID]_[EVIDENCE TYPE]_[NUMBER ID]", where[EVIDENCE TYPE]can be:sentence,cell,header_cell,table_caption,item.context(list of list of str): List (for each element ID incontent) of a list of Wikipedia elements that are automatically associated with that element ID and serve as context. This includes an article's title, relevant sections (the section and sub-section(s) the element is located in), and for cells the closest row and column header (multiple row/column headers if they follow each other).
annotator_operations(list of dict): List of operations an annotator used to find the evidence and reach a verdict, given the claim. Each element in the list is a dictionary with the fields:operation(str): Operation name. Any of the following:start,finish: Annotation started/finished. The value is the name of the operation.search: Annotator used the Wikipedia search function. The value is the entered search term or the term selected from the automatic suggestions. If the annotator did not select any of the suggestions but instead went into advanced search, the term is prefixed with "contains...".hyperlink: Annotator clicked on a hyperlink in the page. The value is the anchor text of the hyperlink.Now on: The page the annotator has landed after a search or a hyperlink click. The value is the PAGE ID.Page search: Annotator search on a page. The value is the search term.page-search-reset: Annotator cleared the search box. The value is the name of the operation.Highlighting,Highlighting deleted: Annotator selected/unselected an element on the page. The value isELEMENT ID.back-button-clicked: Annotator pressed the back button. The value is the name of the operation.
value(str): Value associated with the operation.time(float): Time in seconds from the start of the annotation.
expected_challenge(str): The challenge the claim generator selected will be faced when verifying the claim, one out of the following:Numerical Reasoning,Multi-hop Reasoning,Entity Disambiguation,Combining Tables and Text,Search terms not in claim,Other.challenge(str): Main challenge to verify the claim, one out of the following:Numerical Reasoning,Multi-hop Reasoning,Entity Disambiguation,Combining Tables and Text,Search terms not in claim,Other.
Data Splits
| train | validation | test | |
|---|---|---|---|
| Number of examples | 71291 | 7890 | 7845 |
Dataset Creation
Curation Rationale
Source Data
Initial Data Collection and Normalization
Who are the source language producers?
Annotations
Annotation process
Who are the annotators?
Personal and Sensitive Information
Considerations for Using the Data
Social Impact of Dataset
Discussion of Biases
Other Known Limitations
Additional Information
Dataset Curators
Licensing Information
These data annotations incorporate material from Wikipedia, which is licensed pursuant to the Wikipedia Copyright Policy. These annotations are made available under the license terms described on the applicable Wikipedia article pages, or, where Wikipedia license terms are unavailable, under the Creative Commons Attribution-ShareAlike License (version 3.0), available at http://creativecommons.org/licenses/by-sa/3.0/ (collectively, the “License Termsâ€). You may not use these files except in compliance with the applicable License Terms.
Citation Information
If you use this dataset, please cite:
@inproceedings{Aly21Feverous,
author = {Aly, Rami and Guo, Zhijiang and Schlichtkrull, Michael Sejr and Thorne, James and Vlachos, Andreas and Christodoulopoulos, Christos and Cocarascu, Oana and Mittal, Arpit},
title = {{FEVEROUS}: Fact Extraction and {VERification} Over Unstructured and Structured information},
eprint={2106.05707},
archivePrefix={arXiv},
primaryClass={cs.CL},
year = {2021}
}
Contributions
Thanks to @albertvillanova for adding this dataset.
- Downloads last month
- 56