152 datasets whose data is: labelled data. Computed from the tables of each dataset's ECHO Data Package.
152 datasets — page 3 of 4.
The italy earthquake aug 2016 event of HumAID: 1,201 tweets, each with its ID, its text and one of 6 humanitarian categories assigned by crowdsourcing workers (579 sympathy and support, 312 rescue volunteering or donation effort, 174 injured or dead people, 54 infrastructure and utility damage, 52 other relevant information, 30 requests or urgent needs), split by upstream into 840 train, 122 dev and 239 test tweets.
The Colorado wildfires event of CrisisLexT26 (Colorado, US): 1,200 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 4,182 tweet IDs that the CrisisLex authors collected for the event, posted between 8 June 2012 and 9 July 2012, with a flag for the 4,172 that fall within the period they analyzed. For informativeness, the workers judged 685 related and informative, 268 related but not informative, 238 not related and 9 not applicable of the labeled tweets. The record of the event gives its category (Natural, Climatological, Wildfire), its location and the keywords used to collect the tweets.
The Queensland Floods event of CrisisLexT26 (Queensland, Australia): 1,200 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,223 tweet IDs that the CrisisLex authors collected for the event, posted between 17 January 2013 and 5 February 2013, with a flag for the 1,219 that fall within the period they analyzed. For informativeness, the workers judged 728 related and informative, 191 related but not informative, 261 not related and 20 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The Australia wildfires event of CrisisLexT26 (New South Wales, Australia): 1,199 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,997 tweet IDs that the CrisisLex authors collected for the event, posted between 12 October 2013 and 3 November 2013, with a flag for the 1,982 that fall within the period they analyzed. For informativeness, the workers judged 704 related and informative, 245 related but not informative, 242 not related and 8 not applicable of the labeled tweets. The record of the event gives its category (Natural, Climatological, Wildfire), its location and the keywords used to collect the tweets.
The Arabic training data of the Task 2 (subjectivity in news articles) of the CLEF 2023 CheckThat! Lab: 1,185 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (905), SUBJ (280). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development and test files); this package holds the training file only.
The Arabic training data of the Task 2 (subjectivity in news articles) of the CLEF 2024 CheckThat! Lab: 1,185 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (905), SUBJ (280). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development, dev-test and test files); this package holds the training file only.
The French training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 1,174 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 Glasgow helicopter crash event of CrisisLexT26 (Glasgow, UK): 1,100 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,563 tweet IDs that the CrisisLex authors collected for the event, posted between 29 November 2013 and 31 December 2013, with a flag for the 2,551 that fall within the period they analyzed. For informativeness, the workers judged 578 related and informative, 340 related but not informative, 177 not related and 5 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Crash), its location and the keywords used to collect the tweets.
The Hindi training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 1,081 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 Guatemala earthquake event of CrisisLexT26 (Guatemala, Guatemala): 1,050 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 3,285 tweet IDs that the CrisisLex authors collected for the event, posted between 6 November 2012 and 27 November 2012, with a flag for the 3,261 that fall within the period they analyzed. For informativeness, the workers judged 754 related and informative, 186 related but not informative, 108 not related and 2 not applicable of the labeled tweets. The record of the event gives its category (Natural, Geophysical, Earthquake), its location and the keywords used to collect the tweets.
The Typhoon Yolanda event of CrisisLexT26 (Phillipines, Phillipines): 1,048 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 39,033 tweet IDs that the CrisisLex authors collected for the event, posted between 3 November 2013 and 1 January 2014, with a flag for the 38,881 that fall within the period they analyzed. For informativeness, the workers judged 765 related and informative, 175 related but not informative, 71 not related and 37 not applicable of the labeled tweets. The record of the event gives its category (Natural, Meteorological, Typhoon), its location and the keywords used to collect the tweets.
The LA Airport Shootings event of CrisisLexT26 (Los Angeles, US): 1,032 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,746 tweet IDs that the CrisisLex authors collected for the event, posted between 1 November 2013 and 14 November 2013, with a flag for the 2,730 that fall within the period they analyzed. For informativeness, the workers judged 682 related and informative, 230 related but not informative, 87 not related and 33 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Intentional, Shootings), its location and the keywords used to collect the tweets.
The Italy earthquakes event of CrisisLexT26 (Modena, Italy): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 7,404 tweet IDs that the CrisisLex authors collected for the event, posted between 18 May 2012 and 21 June 2012, with a flag for the 7,351 that fall within the period they analyzed. For informativeness, the workers judged 627 related and informative, 313 related but not informative, 50 not related and 10 not applicable of the labeled tweets. The record of the event gives its category (Natural, Geophysical, Earthquake), its location and the keywords used to collect the tweets.
The Philipinnes Floods event of CrisisLexT26 (Philipinnes, Philipinnes): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,950 tweet IDs that the CrisisLex authors collected for the event, posted between 1 August 2012 and 15 August 2012, with a flag for the 2,917 that fall within the period they analyzed. For informativeness, the workers judged 761 related and informative, 145 related but not informative, 88 not related and 6 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The Typhoon Pablo event of CrisisLexT26 (Phillipines, Phillipines): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,977 tweet IDs that the CrisisLex authors collected for the event, posted between 24 November 2012 and 16 December 2012, with a flag for the 1,944 that fall within the period they analyzed. For informativeness, the workers judged 678 related and informative, 229 related but not informative, 68 not related and 25 not applicable of the labeled tweets. The record of the event gives its category (Natural, Meteorological, Typhoon), its location and the keywords used to collect the tweets.
The Venezuela refinery explosion event of CrisisLexT26 (Falcon, Venezuela): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,845 tweet IDs that the CrisisLex authors collected for the event, posted between 24 August 2012 and 11 September 2012, with a flag for the 2,736 that fall within the period they analyzed. For informativeness, the workers judged 546 related and informative, 393 related but not informative, 58 not related and 3 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Explosion), its location and the keywords used to collect the tweets.
The Alberta Floods event of CrisisLexT26 (Alberta, Canada): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 5,887 tweet IDs that the CrisisLex authors collected for the event, posted between 20 June 2013 and 16 July 2013, with a flag for the 5,789 that fall within the period they analyzed. For informativeness, the workers judged 685 related and informative, 298 related but not informative, 14 not related and 3 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The Bohol earthquake event of CrisisLexT26 (Phillipines, Phillipines): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,214 tweet IDs that the CrisisLex authors collected for the event, posted between 14 October 2013 and 25 October 2013, all of which fall within the period they analyzed. For informativeness, the workers judged 425 related and informative, 544 related but not informative and 31 not related of the labeled tweets. The record of the event gives its category (Natural, Geophysical, Earthquake), its location and the keywords used to collect the tweets.
The Boston Bombings event of CrisisLexT26 (Boston, US): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 157,636 tweet IDs that the CrisisLex authors collected for the event, posted between 15 April 2013 and 19 June 2013, with a flag for the 157,454 that fall within the period they analyzed. For informativeness, the workers judged 417 related and informative, 512 related but not informative, 19 not related and 52 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Intentional, Bombings), its location and the keywords used to collect the tweets.
The Brazil nightclub fire event of CrisisLexT26 (Santa Maria, Brazil): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 4,787 tweet IDs that the CrisisLex authors collected for the event, posted between 27 January 2013 and 12 February 2013, all of which fall within the period they analyzed. For informativeness, the workers judged 442 related and informative, 510 related but not informative, 43 not related and 5 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Fire), its location and the keywords used to collect the tweets.
The Colorado Floods event of CrisisLexT26 (Colorado, US): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,778 tweet IDs that the CrisisLex authors collected for the event, posted between 8 September 2013 and 1 October 2013, with a flag for the 1,752 that fall within the period they analyzed. For informativeness, the workers judged 768 related and informative, 157 related but not informative, 70 not related and 5 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The Lac-Megantic train crash event of CrisisLexT26 (Lac-Megantic, Quebec, Canada): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,497 tweet IDs that the CrisisLex authors collected for the event, posted between 6 July 2013 and 26 July 2013, with a flag for the 2,342 that fall within the period they analyzed. For informativeness, the workers judged 559 related and informative, 407 related but not informative, 31 not related and 3 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Derailment), its location and the keywords used to collect the tweets.
The Manila Floods event of CrisisLexT26 (Manila, Phillipines): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 2,033 tweet IDs that the CrisisLex authors collected for the event, posted between 17 August 2013 and 28 August 2013, with a flag for the 2,032 that fall within the period they analyzed. For informativeness, the workers judged 628 related and informative, 293 related but not informative, 47 not related and 32 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The NYC train crash event of CrisisLexT26 (New York City, US): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,084 tweet IDs that the CrisisLex authors collected for the event, posted between 1 December 2013 and 12 December 2013, with a flag for the 1,066 that fall within the period they analyzed. For informativeness, the workers judged 904 related and informative, 95 related but not informative and 1 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Derailment), its location and the keywords used to collect the tweets.
The Sardinia Floods event of CrisisLexT26 (Sardinia, Italy): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 1,144 tweet IDs that the CrisisLex authors collected for the event, posted between 16 November 2013 and 29 November 2013, with a flag for the 1,143 that fall within the period they analyzed. For informativeness, the workers judged 631 related and informative, 295 related but not informative, 68 not related and 6 not applicable of the labeled tweets. The record of the event gives its category (Natural, Hydrological, Floods), its location and the keywords used to collect the tweets.
The Singapore Haze event of CrisisLexT26 (Singapore, Singapore): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 3,664 tweet IDs that the CrisisLex authors collected for the event, posted between 14 June 2013 and 11 July 2013, with a flag for the 3,639 that fall within the period they analyzed. For informativeness, the workers judged 461 related and informative, 472 related but not informative, 46 not related and 21 not applicable of the labeled tweets. The record of the event gives its category (Mixed, Others, Haze), its location and the keywords used to collect the tweets.
The Spain train crash event of CrisisLexT26 (Santiago de compostela, Spain): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 3,681 tweet IDs that the CrisisLex authors collected for the event, posted between 24 July 2013 and 7 August 2013, all of which fall within the period they analyzed. For informativeness, the workers judged 729 related and informative, 262 related but not informative, 6 not related and 3 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Derailment), its location and the keywords used to collect the tweets.
The West Texas Explosion event of CrisisLexT26 (West/Waco Texas, US): 1,000 tweets about it, each labeled by crowdsourcing workers for informativeness, information type and information source, and the list of 14,599 tweet IDs that the CrisisLex authors collected for the event, posted between 17 April 2013 and 18 May 2013, with a flag for the 14,505 that fall within the period they analyzed. For informativeness, the workers judged 472 related and informative, 439 related but not informative, 65 not related and 24 not applicable of the labeled tweets. The record of the event gives its category (Human-induced, Unintentional, Explosion), its location and the keywords used to collect the tweets.
Weather sentiment from CrowdFlower (Figure Eight): 1,000 weather-related tweets, each graded by 20 contributors for the emotion the author expresses specifically about the weather (271 Negative, 261 Neutral / author is just sharing information, 235 Tweet not related to weather condition, 231 Positive, 2 I can't tell), plus the follow-up job in which 10 contributors judged whether the crowd's grade was correct, for the same 1,000 tweets (answers: 992 Yes, 5 No, 3 I can't tell). Each tweet has an ID, its text, the emotion and the confidence upstream computed for it; the evaluation file's judgments are in their own table.
The English training data of the Task 2 (subjectivity in news articles) of the CLEF 2023 CheckThat! Lab: 830 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (532), SUBJ (298). Each row has a sentence ID, the sentence, the label and a solved_conflict flag. The upstream data folder also holds other files of the task (development and test files); this package holds the training file only.
The English training data of the Task 2 (subjectivity in news articles) of the CLEF 2024 CheckThat! Lab: 830 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (532), SUBJ (298). Each row has a sentence ID, the sentence, the label and a solved_conflict flag. The upstream data folder also holds other files of the task (development, dev-test and test files); this package holds the training file only.
The English training data of the Task 1 (subjectivity in news articles) of the CLEF 2025 CheckThat! Lab: 830 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (532), SUBJ (298). Each row has a sentence ID, the sentence, the label and a solved_conflict flag. The upstream data folder also holds other files of the task (development and dev-test files); this package holds the training file only.
The Dutch training data of the Task 2 (subjectivity in news articles) of the CLEF 2023 CheckThat! Lab: 800 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (489), SUBJ (311). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development and test files); this package holds the training file only.
The German training data of the Task 2 (subjectivity in news articles) of the CLEF 2023 CheckThat! Lab: 800 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (492), SUBJ (308). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development and test files); this package holds the training file only.
The Turkish training data of the Task 2 (subjectivity in news articles) of the CLEF 2023 CheckThat! Lab: 800 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (422), SUBJ (378). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development and test files); this package holds the training file only.
The German training data of the Task 2 (subjectivity in news articles) of the CLEF 2024 CheckThat! Lab: 800 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (492), SUBJ (308). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development, dev-test and test files); this package holds the training file only.
The German training data of the Task 1 (subjectivity in news articles) of the CLEF 2025 CheckThat! Lab: 800 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (492), SUBJ (308). 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 maryland floods 2018 event of HumAID: 742 tweets, each with its ID, its text and one of 8 humanitarian categories assigned by crowdsourcing workers (140 missing or found people, 137 other relevant information, 110 sympathy and support, 79 infrastructure and utility damage, 77 not humanitarian, 73 rescue volunteering or donation effort, 70 caution and advice, 56 injured or dead people), split by upstream into 519 train, 75 dev and 148 test tweets.
The Bulgarian training data of the Task 2 (subjectivity in news articles) of the CLEF 2024 CheckThat! Lab: 729 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (406), SUBJ (323). Each row has a sentence ID, the sentence and the label. The upstream data folder also holds other files of the task (development, dev-test and test files); this package holds the training file only.
The Bulgarian training data of the Task 1 (subjectivity in news articles) of the CLEF 2025 CheckThat! Lab: 729 sentences from news articles, each labelled subjective (SUBJ) or objective (OBJ). The label counts are OBJ (406), SUBJ (323). 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 srilanka floods 2017 event of HumAID: 560 tweets, each with its ID, its text and one of 8 humanitarian categories assigned by crowdsourcing workers (319 rescue volunteering or donation effort, 56 other relevant information, 46 injured or dead people, 40 sympathy and support, 34 requests or urgent needs, 28 caution and advice, 20 not humanitarian, 17 infrastructure and utility damage), split by upstream into 392 train, 57 dev and 111 test tweets.
The Modern Standard Arabic training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 540 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 Weibo part of M3: 529 memes of M3 (image, OCR text, post text, label and rationale), 124 labeled hate and 405 normal. Hate categories: violence (86), politics (16), international relations (14), gender (12), race (10), religion (4), health status (2), public health (1).
The X (Twitter) part of M3: 526 memes of M3 (image, OCR text, post text, label and rationale), 132 labeled hate and 394 normal. Hate categories: politics (81), violence (33), gender (16), religion (16), international relations (14), public health (6), race (5), health status (3).
The Arabic training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 470 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 Punjabi training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 445 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 German training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 386 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 Thai training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 244 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 Polish training data of Task 2 (claim normalization) of the CLEF 2025 CheckThat! Lab: 163 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 Wikipedia hoax and non-hoax articles of Kumar, West and Leskovec (WWW 2016): 76 wikitext sources of hoax articles and 64 of non-hoax articles, plus 64 and 64 versions rendered as Wikipedia pages. Upstream describes 64 hoax articles that were patrolled, survived at least a month and had at least 100 page views a day, and 64 matching non-hoax articles created the same day. Each row is one file, with its path in the archive and its full text; the label (hoax or non-hoax) is in the annotations table.
Twitter (X)
Weibo
Wikipedia