The content of proofassistants.stackexchange.com (https://proofassistants.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 7,911 rows: 2,733 posts (942 questions, 1,303 answers, 241 tag wiki excerpts, 241 tag wikis and 6 posts of type 6) and 5,178 comments, written between 8 February 2022 and 31 March 2024. Beside it are 3,372 user accounts, 15,663 votes, 5,872 badges awarded, 241 tags, 8,901 post revisions and 251 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 12 rows.
Every submission and comment that the Voat dataset of Mekacher and Papasavva holds for the Voat subverse v/BloodOfEurope: 2,709 submissions and 5,234 comments written between 7 November 2018 and 30 July 2020, with the author's user name, upvote and downvote counts, the submission each comment belongs to and each comment's depth in its thread. Voat (voat.co) was a news aggregator in the style of Reddit that shut down on 25 December 2020; its communities were called subverses. The package also holds 843 user profiles (registration date and the subverses the user moderated or created) of the users who wrote in this subverse or moderate or own it, and the profile of the subverse (description, subscriber count, creation date).
The kerala floods 2018 event of HumAID: 7,984 tweets, each with its ID, its text and one of 9 humanitarian categories assigned by crowdsourcing workers (4,294 rescue volunteering or donation effort, 955 other relevant information, 835 sympathy and support, 590 requests or urgent needs, 456 not humanitarian, 363 injured or dead people, 296 infrastructure and utility damage, 139 caution and advice, 56 displaced people and evacuations), split by upstream into 5,588 train, 814 dev and 1,582 test tweets.
Every post and comment collected from the Voat community v/KotakuInAction between 2 January 2015 and 22 December 2020, as published in the MADOC dataset (Multi-Platform Aggregated Dataset of Online Communities). The package holds 7,853 interactions (4,857 comments, 2,996 posts) by 2,051 pseudonymised users. Each interaction has its text, the ID of the post or comment it replies to, and four machine-computed scores: VADER sentiment, TextBlob sentiment, TextBlob subjectivity and a ToxiGen toxicity score. MADOC publishes one file per community and platform; this package is the voat_KotakuInAction_madoc.parquet file.
The hurricane dorian 2019 event of HumAID: 7,613 tweets, each with its ID, its text and one of 9 humanitarian categories assigned by crowdsourcing workers (1,444 other relevant information, 1,369 caution and advice, 1,083 sympathy and support, 987 rescue volunteering or donation effort, 874 not humanitarian, 815 infrastructure and utility damage, 802 displaced people and evacuations, 179 requests or urgent needs, 60 injured or dead people), split by upstream into 5,329 train, 776 dev and 1,508 test tweets.
The california wildfires 2018 event of HumAID: 7,376 tweets, each with its ID, its text and one of 10 humanitarian categories assigned by crowdsourcing workers (1,946 injured or dead people, 1,415 rescue volunteering or donation effort, 1,318 not humanitarian, 1,038 other relevant information, 472 sympathy and support, 422 infrastructure and utility damage, 368 displaced people and evacuations, 179 missing or found people, 139 caution and advice, 79 requests or urgent needs), split by upstream into 5,163 train, 752 dev and 1,461 test tweets.
The hurricane maria 2017 event of HumAID: 7,278 tweets, each with its ID, its text and one of 9 humanitarian categories assigned by crowdsourcing workers (1,977 rescue volunteering or donation effort, 1,568 other relevant information, 1,427 infrastructure and utility damage, 711 requests or urgent needs, 672 sympathy and support, 302 injured or dead people, 270 not humanitarian, 220 caution and advice, 131 displaced people and evacuations), split by upstream into 5,094 train, 742 dev and 1,442 test tweets.
Every submission and comment that the Voat dataset of Mekacher and Papasavva holds for the Voat subverse v/Dreamworlds: 3,366 submissions and 3,782 comments written between 28 March 2015 and 22 December 2020, with the author's user name, upvote and downvote counts, the submission each comment belongs to and each comment's depth in its thread. Voat (voat.co) was a news aggregator in the style of Reddit that shut down on 25 December 2020; its communities were called subverses. The package also holds 176 user profiles (registration date and the subverses the user moderated or created) of the users who wrote in this subverse or moderate or own it, and the profile of the subverse (description, subscriber count, creation date).
The Reddit part of the dataset of 'Who Let The Trolls Out? Towards Understanding State-Sponsored Trolls' (Zannettou, Caulfield, Setzer, Sirivianos, Stringhini and Blackburn): 7,097 comments and 14,224 submissions by 335 Reddit accounts that the authors identified as belonging to the Russian Internet Research Agency, as the Reddit API returned them, posted between 2 September 2011 and 9 April 2018; the 943 account records (name, creation time, karma and profile fields) of 940 different troll accounts; and the list of 944 account names. The models and the Twitter data of the same Zenodo record are not part of this package.
The content of stellar.stackexchange.com (https://stellar.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 5,295 rows: 3,268 posts (1,463 questions, 1,674 answers, 65 tag wiki excerpts, 65 tag wikis and 1 post of type 6) and 2,027 comments, written between 17 January 2018 and 28 March 2024. Beside it are 7,156 user accounts, 9,253 votes, 7,652 badges awarded, 115 tags, 10,071 post revisions and 149 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 12 rows.
Every submission and comment that the Voat dataset of Mekacher and Papasavva holds for the Voat subverse v/newsandpolitics: 3,465 submissions and 3,143 comments written between 19 September 2015 and 24 December 2020, with the author's user name, upvote and downvote counts, the submission each comment belongs to and each comment's depth in its thread. Voat (voat.co) was a news aggregator in the style of Reddit that shut down on 25 December 2020; its communities were called subverses. The package also holds 88 user profiles (registration date and the subverses the user moderated or created) of the users who wrote in this subverse or moderate or own it, and the profile of the subverse (description, subscriber count, creation date).
Every submission and comment that the Voat dataset of Mekacher and Papasavva holds for the Voat subverse v/Contact: 1,516 submissions and 4,627 comments written between 27 May 2015 and 24 December 2020, with the author's user name, upvote and downvote counts, the submission each comment belongs to and each comment's depth in its thread. Voat (voat.co) was a news aggregator in the style of Reddit that shut down on 25 December 2020; its communities were called subverses. The package also holds 364 user profiles (registration date and the subverses the user moderated or created) of the users who wrote in this subverse or moderate or own it, and the profile of the subverse (description, subscriber count, creation date).
The content of vegetarianism.stackexchange.com (https://vegetarianism.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 5,565 rows: 2,409 posts (760 questions, 1,497 answers, 73 tag wiki excerpts, 73 tag wikis and 6 posts of type 6) and 3,156 comments, written between 31 January 2017 and 31 March 2024. Beside it are 5,075 user accounts, 16,039 votes, 7,661 badges awarded, 117 tags, 8,263 post revisions and 314 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 32 rows.
The hurricane florence 2018 event of HumAID: 6,264 tweets, each with its ID, its text and one of 9 humanitarian categories assigned by crowdsourcing workers (1,478 rescue volunteering or donation effort, 1,310 caution and advice, 1,060 not humanitarian, 637 displaced people and evacuations, 636 other relevant information, 472 sympathy and support, 320 infrastructure and utility damage, 297 injured or dead people, 54 requests or urgent needs), split by upstream into 4,384 train, 639 dev and 1,241 test tweets.
Tweets of the Twitter accounts of the Fortune 100 companies of 2020 in the weeks around the death of George Floyd, the data of Kevin McElwee's 'The Corporate Response to #BlackLivesMatter on Twitter': 140,279 tweets from 91 company accounts as returned by the Twitter API (the Twitter API returns at most an account's latest 3,200; 12 March 2009 to 27 July 2020), the list of the 100 companies with their Twitter handle, rank and sector, and manual labels for the 6,184 tweets sent between 25 May and 25 July 2020: whether each tweet addresses racial inequity (461 do) and, for those, whether it contains the words Black Lives Matter, mentions Juneteenth or mentions donations. A second file flags 461 tweets for BLM, Juneteenth, Money and Formal Statement.
The content of iota.stackexchange.com (https://iota.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 4,895 rows: 2,464 posts (1,074 questions, 1,283 answers, 51 tag wiki excerpts, 51 tag wikis and 5 posts of type 6) and 2,431 comments, written between 28 November 2017 and 1 March 2024. Beside it are 6,040 user accounts, 10,698 votes, 7,077 badges awarded, 159 tags, 7,869 post revisions and 310 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 36 rows.
The content of drones.stackexchange.com (https://drones.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 5,197 rows: 2,612 posts (908 questions, 1,277 answers, 208 tag wiki excerpts, 208 tag wikis, 8 posts of type 6 and 3 posts of type 7) and 2,585 comments, written between 14 April 2020 and 21 March 2024. Beside it are 5,251 user accounts, 9,983 votes, 6,281 badges awarded, 224 tags, 7,829 post revisions and 109 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 10 rows.
Posts of 139 bloggers from the Blog Authorship Corpus collected by Jonathan Schler, Moshe Koppel, Shlomo Argamon and James Pennebaker from blogger.com: the bloggers whose industry is Religion, aged 13 to 48 (43 female, 96 male). The package holds 5,235 posts, each with its date and text, and one record per blogger with the gender, age, industry and astrological sign the blogger gave. The posts are dated from 1 January 1999 to 10 November 2004. The corpus is one file per blogger; this package holds the files of one slice of it.
The content of conlang.stackexchange.com (https://conlang.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 4,662 rows: 1,948 posts (603 questions, 1,184 answers, 78 tag wiki excerpts, 78 tag wikis and 5 posts of type 6) and 2,714 comments, written between 6 February 2018 and 28 March 2024. Beside it are 3,756 user accounts, 11,456 votes, 5,415 badges awarded, 111 tags, 5,214 post revisions and 85 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 9 rows.
5,171 short social media posts about NFL teams, written between 2024-11-23 and 2024-12-17, scraped with a Selenium web scraper by Cameron C. (Kaggle owner cammyc), each with the author's display name, the team the post was collected for and a sentiment label (neutral (2,235), negative (1,757), positive (1,161) among well-formed rows) predicted by a fine-tuned RoBERTa model, with its confidence and raw class probabilities.
2015 New Year's resolutions from CrowdFlower: 5,011 tweets, a Twitter sentiment analysis of users' 2015 New Year's resolutions that, per the publisher, contains demographic and geographical data of the users and resolution categorizations. Each tweet has a resolution topic (115 distinct values) and one of 10 resolution categories (1,781 Personal Growth, 1,010 Humor, 840 Health & Fitness, 467 Recreation & Leisure, 351 Family/Friends/Relationships, 176 Finance, 126 Career, 89 Education/Training, 87 Time Management/Organization, 84 Philanthropic). Each row also has the tweet text, the author's screen name and a gender value (2,518 male, 2,493 female), the retweet count, time, free-text location, state, region and time zone; 227 tweets carry coordinates.
The Amharic part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Amharic the dataset card names the Amharic Hate Speech dataset as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 5,000 of the texts are distinct; the rest occur more than once.
The Arabic part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Arabic the dataset card names OSACT4 as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,991 of the texts are distinct; the rest occur more than once.
The German part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For German the dataset card names the DeTox dataset and GermEval 2018/2021 as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,585 of the texts are distinct; the rest occur more than once.
The English part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For English the dataset card names Jigsaw and the Unitary AI Toxicity Dataset as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,900 of the texts are distinct; the rest occur more than once.
The Spanish part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Spanish the dataset card names the CLANDESTINO Spanish toxic language dataset as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 5,000 of the texts are distinct; the rest occur more than once.
The French part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). The dataset card lists no source for French. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,774 of the texts are distinct; the rest occur more than once.
The Hindi part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Hindi the dataset card names the Hostility Detection Dataset in Hindi and the HASOC 2019 track overview as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,963 of the texts are distinct; the rest occur more than once.
The Italian part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Italian the dataset card names AMI, HODI 2023 and the Jigsaw Multilingual Toxic Comment competition as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,975 of the texts are distinct; the rest occur more than once.
The Japanese part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Japanese the dataset card names posts from the 2channel-derived open2ch forum, filtered with the Perspective API as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,908 of the texts are distinct; the rest occur more than once.
The Russian part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Russian the dataset card names two Kaggle datasets, "Russian Language Toxic Comments" and "Toxic Russian Comments" as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,978 of the texts are distinct; the rest occur more than once.
The Tatar part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Tatar the dataset card names the textdetox authors' own data as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 5,000 of the texts are distinct; the rest occur more than once.
The Ukrainian part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). For Ukrainian the dataset card names the dataset of Dementieva, Khylenko, Babakov and Groh (WOAH 2024), hosted on Hugging Face as the source. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,999 of the texts are distinct; the rest occur more than once.
The Chinese part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 5,000 short texts, each labelled toxic or non-toxic (2,500 toxic, 2,500 non-toxic). The dataset card lists no source for Chinese. The file has two columns, the text and the binary label, and no identifiers, dates or author information. 5,000 of the texts are distinct; the rest occur more than once.
The content of moderators.stackexchange.com (https://moderators.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 3,870 rows: 1,992 posts (582 questions, 1,296 answers, 55 tag wiki excerpts, 55 tag wikis, 1 post of type 6 and 3 posts of type 7) and 1,878 comments, written between 29 July 2014 and 31 March 2024. Beside it are 7,184 user accounts, 16,555 votes, 9,343 badges awarded, 74 tags, 5,471 post revisions and 135 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 10 rows.
The Hinglish (Hindi written in Latin script) part of the TextDetox multilingual toxicity dataset, compiled for the PAN 2024 and PAN 2025 TextDetox shared tasks: 4,363 short texts, each labelled toxic or non-toxic (2,200 toxic, 2,163 non-toxic). The dataset card lists no source for Hinglish (Hindi written in Latin script). The file has two columns, the text and the binary label, and no identifiers, dates or author information. 4,237 of the texts are distinct; the rest occur more than once.
Posts of 112 bloggers from the Blog Authorship Corpus collected by Jonathan Schler, Moshe Koppel, Shlomo Argamon and James Pennebaker from blogger.com: the bloggers whose industry is Banking, aged 13 to 48 (47 female, 65 male). The package holds 4,049 posts, each with its date and text, and one record per blogger with the gender, age, industry and astrological sign the blogger gave. The posts are dated from 22 January 2001 to 24 August 2004. The corpus is one file per blogger; this package holds the files of one slice of it.
The cyclone idai 2019 event of HumAID: 3,933 tweets, each with its ID, its text and one of 10 humanitarian categories assigned by crowdsourcing workers (1,869 rescue volunteering or donation effort, 482 sympathy and support, 433 injured or dead people, 407 other relevant information, 354 infrastructure and utility damage, 143 requests or urgent needs, 89 caution and advice, 80 not humanitarian, 57 displaced people and evacuations, 19 missing or found people), split by upstream into 2,753 train, 401 dev and 779 test tweets.
The English part of OffensEval 2020 (SemEval-2020 Task 12): 3,887 tweets labeled for offensive language, 1,080 OFF (offensive) and 2,807 NOT (not offensive) for sub-task A, 0 in the training file and 3,887 in the test file (labeled by the shared task's separate label file). The English training data is not part of the files.
Every post and comment collected from the Voat community v/MensRights between 25 June 2014 and 30 November 2020, as published in the MADOC dataset (Multi-Platform Aggregated Dataset of Online Communities). The package holds 3,735 interactions (2,010 comments, 1,725 posts) by 1,125 pseudonymised users. Each interaction has its text, the ID of the post or comment it replies to, and four machine-computed scores: VADER sentiment, TextBlob sentiment, TextBlob subjectivity and a ToxiGen toxicity score. MADOC publishes one file per community and platform; this package is the voat_MensRights_madoc.parquet file.
The content of bioacoustics.stackexchange.com (https://bioacoustics.stackexchange.com/), a Stack Exchange question-and-answer site, as published in the Stack Exchange Data Dump of April 2024 on the Internet Archive. The posts table holds 3,097 rows: 1,598 posts (480 questions, 1,074 answers, 17 tag wiki excerpts, 17 tag wikis and 10 posts of type 6) and 1,499 comments, written between 22 June 2022 and 30 March 2024. Beside it are 1,998 user accounts, 10,217 votes, 3,351 badges awarded, 211 tags, 3,988 post revisions and 97 links between posts. Every answer keeps the question it answers and every comment the post it was written on, so the threads can be rebuilt. The Internet Archive item describes the dump as anonymised; the votes table names the voter in only 8 rows.
The Spanish training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 3,458 social media posts, each paired with a normalized claim, a short simplified statement of the claim the post makes. Each row has the post and its normalized claim. The task asks systems to generate the normalized claim from the noisy post. The upstream repository also holds development and test files for this task; this package holds the training file only.
The #petpeeve dataset of William Yang Wang and Diyi Yang (EMNLP 2015): 3,375 tweets that use the hashtag #petpeeve, each assigned one of 60 categories of annoying behaviour (the most frequent are others, 585, and appearance, 463). The data come as a training file with 2,025 tweets and a test file with 1,350. Each row is the category and the tweet text; the files carry no tweet IDs, dates or user names.
The Danish part of OffensEval 2020 (SemEval-2020 Task 12): 3,290 tweets labeled for offensive language, 425 OFF (offensive) and 2,865 NOT (not offensive) for sub-task A, 2,961 in the training file and 329 in the test file (labeled by the shared task's separate label file).
Every post and comment collected from the Voat community v/CringeAnarchy between 20 June 2015 and 22 December 2020, as published in the MADOC dataset (Multi-Platform Aggregated Dataset of Online Communities). The package holds 3,030 interactions (2,091 comments, 939 posts) by 919 pseudonymised users. Each interaction has its text, the ID of the post or comment it replies to, and four machine-computed scores: VADER sentiment, TextBlob sentiment, TextBlob subjectivity and a ToxiGen toxicity score. MADOC publishes one file per community and platform; this package is the voat_CringeAnarchy_madoc.parquet file.
Elon Musk's tweets from 4 June 2010 to 5 April 2017: 2,819 tweets and replies, each with its tweet ID, creation time and text, as compiled by the data.world user adamhelsinger ("Elon Musk Tweets-Until 4/6/17"). The description of that data.world dataset says it holds the number from the tweet's address as the ID, the date and time the tweet was created, and the tweets and mentions. The package has no other fields: no author column, no counts and no replies structure.
M3 (Multi-platform, Multi-lingual, and Multimodal Meme), a benchmark for multimodal hate speech detection by Rui Xing, Qi Chai, Jie Ma, Jing Tao, Pinghui Wang, Shuming Zhang, Xinping Wang and Hao Wang: 2,455 memes, each an image with the text of the social media post it was shared in, collected from X (Twitter, 526 memes), 4chan (1,400) and Weibo (529). Each meme has the text read from its image (OCR), the post text, a label hate (1,318) or normal (1,137), the hate categories it falls in, and a short rationale. The hate categories are politics (348), race (334), violence (297), international relations (180), gender (172), religion (108), health status (106), public health (85); the 1,137 normal memes have the category none. One table of posts serves all three platforms; the column collection_member names the platform, so a single platform is a filter of the table.
The Arabic training data of the Task 1 (subjectivity in news articles) of the CLEF 2025 CheckThat! Lab: 2,446 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (1,391), SUBJ (1,055). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development and dev-test files); this package holds the training file only.
The canada wildfires 2016 event of HumAID: 2,242 tweets, each with its ID, its text and one of 8 humanitarian categories assigned by crowdsourcing workers (934 rescue volunteering or donation effort, 380 displaced people and evacuations, 311 other relevant information, 251 infrastructure and utility damage, 161 sympathy and support, 106 caution and advice, 79 not humanitarian, 20 requests or urgent needs), split by upstream into 1,569 train, 228 dev and 445 test tweets.
The kaikoura earthquake 2016 event of HumAID: 2,195 tweets, each with its ID, its text and one of 9 humanitarian categories assigned by crowdsourcing workers (493 caution and advice, 432 sympathy and support, 312 infrastructure and utility damage, 311 other relevant information, 224 not humanitarian, 207 rescue volunteering or donation effort, 105 injured or dead people, 87 displaced people and evacuations, 24 requests or urgent needs), split by upstream into 1,536 train, 224 dev and 435 test tweets.
Stack Exchange
Voat
Twitter (X)
Reddit
Blogger
4chan
Weibo