A TextDetox 2024 shared task (Spanish) dataset of 5,000 Spanish posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Spanish provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (French) dataset of 5,000 French posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox French provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Hindi) dataset of 5,000 Hindi posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Hindi provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Italian) dataset of 5,000 Italian posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Italian provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Japanese) dataset of 5,000 Japanese posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Japanese provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Russian) dataset of 5,000 Russian posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Russian provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Tatar) dataset of 5,000 Tatar posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Tatar provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Ukrainian) dataset of 5,000 Ukrainian posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Ukrainian provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Chinese) dataset of 5,000 Chinese posts (2,500 toxic, 2,500 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Chinese provides 5,000 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A TextDetox 2024 shared task (Hindglish) dataset of 4,363 Hindglish posts (2,200 toxic, 2,163 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Hindglish provides 4,363 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A Twitter (OffensEval 2020 — English) dataset of 3,887 English tweets (1,080 offensive, 2,807 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 English — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A Voat (Reddit alternative, 2013–2020) dataset of 3,366 submissions, 595 comments, 174 users, 218 collaboration edges, collected by Mekacher & Papasavva (ICWSM 2022) via full platform crawl. Each record contains user nodes (registration date, moderation status), directed weighted edges (commenter→submitter, weight=upvotes+1), submission titles and bodies, comment text, upvotes, downvotes, depth. Complete dataset for the Voat subverse v/Dreamworlds, covering its full lifetime on the platform (2013–2020). Includes all 3,366 submissions and 595 comments with full text and engagement metadata.. Suitable for social network analysis, community dynamics, content moderation, platform migration studies, online community research.
A Voat (Reddit alternative, 2013–2020) dataset of 1,516 submissions, 2,089 comments, 376 users, 439 collaboration edges, collected by Mekacher & Papasavva (ICWSM 2022) via full platform crawl. Each record contains user nodes (registration date, moderation status), directed weighted edges (commenter→submitter, weight=upvotes+1), submission titles and bodies, comment text, upvotes, downvotes, depth. Complete dataset for the Voat subverse v/Contact, covering its full lifetime on the platform (2013–2020). Includes all 1,516 submissions and 2,089 comments with full text and engagement metadata.. Suitable for social network analysis, community dynamics, content moderation, platform migration studies, online community research.
A Voat (Reddit alternative, 2013–2020) dataset of 3,465 submissions, 117 comments, 73 users, 66 collaboration edges, collected by Mekacher & Papasavva (ICWSM 2022) via full platform crawl. Each record contains user nodes (registration date, moderation status), directed weighted edges (commenter→submitter, weight=upvotes+1), submission titles and bodies, comment text, upvotes, downvotes, depth. Complete dataset for the Voat subverse v/newsandpolitics, covering its full lifetime on the platform (2013–2020). Includes all 3,465 submissions and 117 comments with full text and engagement metadata.. Suitable for social network analysis, community dynamics, content moderation, platform migration studies, online community research.
A HOT Speech (hate/offensive/toxic speech dataset) dataset of 3,481 social media comments with multi-dimensional annotations, collected by SOMAR / ICPSR via Twitter, YouTube, and Reddit APIs (August 2021). Each record contains comment text, hard and soft labels (hateful, offensive, toxic), composite scores, and 21 researcher codes. multi-platform dataset with rich annotations including both MTurk worker labels and researcher codes. Suitable for hate speech detection, offensive language identification, toxicity classification, and political discourse analysis.
A Twitter (OffensEval 2020 — Danish) dataset of 3,290 Danish tweets (425 offensive, 2,865 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 Danish — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A 4chan, Twitter, and Weibo (cross-platform) dataset of 2,455 multimodal posts (1,318 hate, 1,137 non-hate), collected by Mira AI Lab (2024). Each record contains image filename, image OCR text, post text, hate/non-hate binary label, hate category (gender/race/religion/politics/violence/health/etc), and human-annotated reason for classification. Multimodal cross-platform collection hate speech dataset combining image OCR text and post text with fine-grained category labels across 8 hate dimensions — 54% of posts are hate speech with human-annotated reasoning. Suitable for multimodal hate speech detection, cross-platform analysis, hate category classification, image-text hate speech, content moderation research. The package includes the 2,455 meme images (img/) that the records name.
A TextDetox 2024 shared task (Hebrew) dataset of 2,011 Hebrew posts (807 toxic, 1,204 non-toxic), collected by Dementieva et al. for PAN/CLEF TextDetox 2024. Each record contains post text, binary toxicity label (0=non-toxic, 1=toxic). TextDetox Hebrew provides 2,011 balanced binary toxicity annotations — part of a 15-language multilingual benchmark for cross-lingual toxicity detection research. Suitable for toxicity detection, multilingual content moderation, cross-lingual NLP, hate speech classification, text detoxification.
A 4chan dataset of 1,400 multimodal posts (1,062 hate, 338 non-hate), collected by Mira AI Lab (2024). Each record contains image filename, image OCR text, post text, hate/non-hate binary label, hate category (gender/race/religion/politics/violence/health/etc), and human-annotated reason for classification. Multimodal 4chan subset hate speech dataset combining image OCR text and post text with fine-grained category labels across 8 hate dimensions — 76% of posts are hate speech with human-annotated reasoning. Suitable for multimodal hate speech detection, cross-platform analysis, hate category classification, image-text hate speech, content moderation research. The package includes the 1,400 meme images (img/) that the records name.
A NoisyHate (hate speech benchmark) dataset of 1,339 labeled sentences with text perturbations and toxicity scores, collected by NoisyHate Authors via ICWSM 2025. Each record contains sentence text, binary hate label, toxicity scores (toxicity, severe, obscene, identity_attack, insult, threat), identity group indicators, and engagement metrics. The dataset is notable for its benchmark dataset with human-written text perturbations and fine-grained toxicity metrics. Intended for research into hate speech detection, model robustness, adversarial testing, toxicity analysis, and identity-based hate speech research.
A Cross-platform meme corpus (4chan, Twitter, Weibo) dataset of 1,318 hate-speech labeled meme records from 3 platforms, collected by Mira AI Lab via cross-platform meme collection. Each record contains image filename, OCR-extracted image text, post caption text, hate speech label, category list (politics/race/religion/etc.), and human-readable annotation reason per record. M3-Hate is the hate-speech filtered subset of the M3 meme dataset, with 1,318 hateful memes across 4chan (1062), Twitter (132), and Weibo (124) — multimodal with both image OCR text and post captions. Suitable for hate speech detection, multimodal meme analysis, cross-platform content moderation, image-text classification, harmful content research. The package includes the 1,318 meme images (img/) alongside the JSON records.
A Voat (Reddit alternative, 2013–2020) dataset of 321 submissions, 422 comments, 42 users, 63 collaboration edges, collected by Mekacher & Papasavva (ICWSM 2022) via full platform crawl. Each record contains user nodes (registration date, moderation status), directed weighted edges (commenter→submitter, weight=upvotes+1), submission titles and bodies, comment text, upvotes, downvotes, depth. Complete dataset for the Voat subverse v/PizzagateNoMod, covering its full lifetime on the platform (2013–2020). Includes all 321 submissions and 422 comments with full text and engagement metadata.. Suitable for social network analysis, community dynamics, content moderation, platform migration studies, online community research.
A Weibo dataset of 529 multimodal posts (124 hate, 405 non-hate), collected by Mira AI Lab (2024). Each record contains image filename, image OCR text, post text, hate/non-hate binary label, hate category (gender/race/religion/politics/violence/health/etc), and human-annotated reason for classification. Multimodal Weibo subset hate speech dataset combining image OCR text and post text with fine-grained category labels across 8 hate dimensions — 23% of posts are hate speech with human-annotated reasoning. Suitable for multimodal hate speech detection, cross-platform analysis, hate category classification, image-text hate speech, content moderation research. The package includes the 529 meme images (img/) that the records name.
A Twitter/X dataset of 526 multimodal posts (132 hate, 394 non-hate), collected by Mira AI Lab (2024). Each record contains image filename, image OCR text, post text, hate/non-hate binary label, hate category (gender/race/religion/politics/violence/health/etc), and human-annotated reason for classification. Multimodal Twitter subset hate speech dataset combining image OCR text and post text with fine-grained category labels across 8 hate dimensions — 25% of posts are hate speech with human-annotated reasoning. Suitable for multimodal hate speech detection, cross-platform analysis, hate category classification, image-text hate speech, content moderation research. The package includes the 526 meme images (img/) that the records name.
A TikTok (pro-eating disorder content) dataset of 85 videos and photo posts annotated using the EDE clinical framework, collected by Shaveet, Viranda, and Choudhury at Cornell University. Each record contains content type (video/photos), human-written content summary, EDE subscale and diagnostic item annotations, visual and audio element labels, audio tempo/valence/arousal scores, harmonic ratio, and mode. EDTT is the first clinically-grounded dataset of pro-eating-disorder TikTok content, with 47 videos and 38 photo posts annotated using the Eating Disorder Examination framework by Cornell researchers. Suitable for eating disorder content detection, harmful content moderation, multimodal annotation research, TikTok platform analysis, clinical NLP, audio-visual content analysis.
A twitter dataset of tweet IDs from the 2017 Charlottesville protests, collected by Littman via Social Feed Manager (GWU Libraries). Each record contains tweet IDs only — full content requires hydration. Focused crisis dataset from a pivotal far-right rally and counter-protest event. Suitable for crisis informatics, hate speech, political extremism, social media during unrest.
Twitter (X)
Voat
Reddit
YouTube
4chan
Weibo
TikTok