A BlogCatalog (v3) dataset of 10,312 bloggers and 333,983 friendship edges, collected by Agarwal et al. / Tang & Liu via BlogCatalog crawl. Each record contains blogger-blogger friendship edges, user IDs and group memberships. The dataset is notable for its blogosphere friendship network from BlogCatalog, used for familiar stranger detection and latent social dimensions research. Intended for research into social network analysis, community detection, group membership prediction, and blogosphere structure studies.
A News Articles (multiple sources) dataset of 9,900 news articles (5,000 fake, 4,900 real), collected by Vishakh Dapat via Kaggle (2023). Each record contains full article text and binary veracity label (Fake or Real). The dataset is notable for its balanced binary-labeled news corpus with approximately equal proportions of fake and real articles, making it well-suited as a classification benchmark. Intended for research into fake news detection, text classification benchmarking, misinformation topic modeling, and NLP feature engineering.
A Gnutella (August 05, 2002) dataset of 8,846 nodes and 31,839 edges, collected by Ripeanu et al. via Gnutella crawl (2002). Each record contains directed peer-to-peer connections in the Gnutella network. The dataset is notable for its snapshot from August 05, 2002, part of a temporal sequence of 9 snapshots. Intended for research into peer-to-peer network analysis, temporal graph evolution, and network resilience studies.
A Gnutella (August 06, 2002) dataset of 8,717 nodes and 31,525 edges, collected by Ripeanu et al. via Gnutella crawl (2002). Each record contains directed peer-to-peer connections in the Gnutella network. The dataset is notable for its snapshot from August 06, 2002, part of a temporal sequence of 9 snapshots. Intended for research into peer-to-peer network analysis, temporal graph evolution, and network resilience studies.
A Gnutella (August 09, 2002) dataset of 8,114 nodes and 26,013 edges, collected by Ripeanu et al. via Gnutella crawl (2002). Each record contains directed peer-to-peer connections in the Gnutella network. The dataset is notable for its snapshot from August 09, 2002, part of a temporal sequence of 9 snapshots. Intended for research into peer-to-peer network analysis, temporal graph evolution, and network resilience studies.
A Usenet (sample) dataset of 7,361 posts from 801 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 7 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 182 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 53 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 710 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 215 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 273 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A Usenet (sample) dataset of 7,361 posts from 43 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A CLEF CheckThat! 2024 shared task (arabic) dataset of 7,333 sentences with binary check-worthiness labels (30.6% check-worthy), collected by CLEF CheckThat! Lab organizers (2024). Each record contains sentence text, binary check-worthiness label (Yes/No), sentence ID, source document. CLEF 2024 CheckThat! Task 1 (arabic) provides 7,333 sentences from political debates and social media annotated for check-worthiness — whether the claim merits fact-checking. Suitable for check-worthiness detection, fact-checking automation, misinformation research, NLP classification benchmarks, political discourse analysis.
A Chess game records (professional matches) dataset of 7,301 nodes and 60,046 edges, collected by KONECT via chess game database. Each record contains directed: True, weighted: True, signed: True. Directed network of chess game outcomes: edges go from winner to loser, with timestamps. Draws encoded as 0 weight.. Suitable for competition network analysis, ranking algorithms, temporal network analysis.
A MOOC online learning platform dataset of 7,047 students, 97 course content items, and 411,749 student actions, collected by Kumar, Zhang & Leskovec via SNAP (2019). Each record contains action ID, student user ID, course content item ID, relative timestamp, dropout label (0/1), and 4 pre-computed action feature dimensions. Bipartite temporal network of student interactions with MOOC course content — 1.0% dropout rate across 396,633 actions. Includes pre-computed embedding features per action, enabling direct use in temporal graph learning without feature engineering.. Suitable for student dropout prediction, temporal interaction networks, dynamic graph embedding, online learning behavior analysis.
A Rumble (alt-right video platform) dataset of 6,735 Rumble videos from 245 channels (2020-2023), collected by iDRAMA Lab via Rumble API (2024). Each record contains video ID, title, channel name, view/upvote/dislike/comment counts, follower count, video duration, description, named entities, tags, URL. 6,735 Rumble videos with full engagement metrics and NER annotations — one of few open datasets covering Rumble, a major alternative video platform popular with right-wing creators post-YouTube deplatforming. Suitable for alternative video platform research, deplatforming effects, political video discourse, engagement analysis, NER on video metadata.
A Rumble (alt-right video platform) — speech transcriptions dataset of 6,735 lite speech transcriptions from Rumble podcasts (2020-2023), collected by iDRAMA Lab via Whisper ASR (2024). Each record contains video ID, lite transcription segments, named entities (person/org/GPE). 6,735 Rumble podcast transcriptions enabling text-based analysis of spoken political content — rare speech-to-text dataset from an alternative video platform. Suitable for spoken political discourse analysis, NER on speech, alternative platform audio content research, misinformation in video content.
A Reactome biological pathway database dataset of 6,327 nodes and 147,547 edges, collected by Reactome consortium. Each record contains directed: False, weighted: False, signed: False. Protein-protein interaction network from the Reactome pathway database — large-scale biological network.. Suitable for biological network analysis, community detection, protein interaction research.
A Gnutella (August 08, 2002) dataset of 6,301 nodes and 20,777 edges, collected by Ripeanu et al. via Gnutella crawl (2002). Each record contains directed peer-to-peer connections in the Gnutella network. The dataset is notable for its snapshot from August 08, 2002, part of a temporal sequence of 9 snapshots. Intended for research into peer-to-peer network analysis, temporal graph evolution, and network resilience studies.
A Usenet (sample) dataset of 6,111 posts from 2470 newsgroups, collected by OwnedByDanes via Hugging Face. Each record contains post text, newsgroup, date, author, subject. Sample of Usenet posts from 1980–2013, with 5,000 posts per top‑level hierarchy. Covers comp.*, sci.*, rec.*, alt.*, etc.. Suitable for historical discourse analysis, early internet culture, topic evolution, authorship attribution.
A CoAID (COVID-19 misinformation dataset) dataset of 5,975 news articles and social media posts, collected by Cui and Lee via COVID-19 pandemic (2020-2021). Each record contains news articles, fact-check URLs, binary label (0=real, 1=fake), and temporal data. The dataset is notable for its temporally-structured dataset tracking COVID-19 misinformation evolution. Intended for research into COVID-19 misinformation detection, healthcare communication, and temporal social media analysis.
A Bitcoin OTC (peer-to-peer Bitcoin trading platform) dataset of 5,881 users and 35,592 directed weighted trust ratings (32,029 positive, 3,563 negative), collected by Kumar et al. via platform crawl, distributed by SNAP. Each record contains directed weighted signed edges between users: integer rating in [-10, +10] representing trust strength, plus binarized sign (+1/-1) for compatibility with signed network algorithms; Unix timestamps per rating. continuous integer trust weights rather than binary +1/-1 signs, enabling weighted signed network analysis and providing finer-grained trust signal than Epinions or Slashdot while operating in the high-stakes context of peer-to-peer Bitcoin trading. Suitable for weighted signed network analysis, edge weight prediction, trust propagation modeling, fraud detection, reputation system design, and signed graph embedding research.
A NDS-VU signed network repository dataset of 5,882 nodes and 21,492 signed edges (synthetic Bitcoin OTC variant), collected by NDS-VU (Vrije Universiteit Amsterdam). Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Synthetic version of the Bitcoin OTC web-of-trust network with reversed edge polarities, enabling adversarial and robustness testing of signed network algorithms. Suitable for signed network analysis, algorithm benchmarking, robustness testing.
A TextDetox 2024 shared task (Amharic) dataset of 5,000 Amharic 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 Amharic 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 (Arabic) dataset of 5,000 Arabic 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 Arabic 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 (German) dataset of 5,000 German 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 German 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 (English) dataset of 5,000 English 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 English 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 (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 US Western States power grid infrastructure dataset of 4,941 nodes and 6,594 edges, collected by D.J. Watts & S.H. Strogatz. Each record contains directed: False, weighted: False, signed: False. Classic small-world network of the US western power grid — nodes are generators/transformers, edges are transmission lines.. Suitable for infrastructure network analysis, small-world networks, robustness analysis.
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 Bitcoin Alpha (peer-to-peer Bitcoin trading platform) dataset of 3,783 users and 24,186 directed weighted trust ratings (22,650 positive, 1,536 negative), collected by Kumar et al. via platform crawl, distributed by SNAP. Each record contains directed weighted signed edges between users: integer rating in [-10, +10] representing trust strength, plus binarized sign (+1/-1) for compatibility with signed network algorithms; Unix timestamps per rating. continuous integer trust weights rather than binary +1/-1 signs, enabling weighted signed network analysis and providing finer-grained trust signal than Epinions or Slashdot while operating in the high-stakes context of peer-to-peer Bitcoin trading. Suitable for weighted signed network analysis, edge weight prediction, trust propagation modeling, fraud detection, reputation system design, and signed graph embedding research.
A CLEF CheckThat! 2025 Task 2 (spanish) dataset of 3,458 social media post and normalized claim pairs, collected by CLEF CheckThat! Lab organizers (2025). Each record contains social media post text containing a factual claim, normalized canonical claim statement. CLEF 2025 CheckThat! Task 2 (spanish) provides 3,458 pairs of social media posts and their normalized factual claims, enabling claim normalization and canonicalization research. Suitable for claim normalization, fact-checking automation, NLP text generation, multilingual claim extraction, social media fact verification.
A OpenFlights.org airport route database dataset of 3,425 nodes and 37,595 edges, collected by OpenFlights.org. Each record contains directed: True, weighted: True. Directed network of airports connected by flight routes — nodes are airports, edges are direct routes.. Suitable for transportation network analysis, network robustness, hub detection.
A #petpeeve (annoying behaviors) dataset of 3,375 tweets labeled with annoying behavior categories, collected by Wang and Yang via Twitter API (2015). Each record contains tweet text and annoying behavior category label. The dataset is notable for its dataset for automatic categorization of annoying behaviors using lexical and frame-semantic embeddings. Intended for research into social media analysis, text classification, and behavioral categorization.
A CiteSeer dataset of 3,327 nodes, 4,676 edges, 6 classes, collected by Yang et al. (ICML 2016). Each record contains node features, citation edges, class labels. CiteSeer network with bag-of-words features. Suitable for node classification, semi-supervised learning, graph neural networks.
A Human protein interaction network (Vidal lab) dataset of 3,133 nodes and 6,726 edges, collected by M. Vidal et al.. Each record contains directed: False, weighted: False. Human protein-protein interaction network from the Vidal lab — classic biological network benchmark.. Suitable for biological network analysis, protein interaction research.
A Cora dataset of 2,708 nodes, 5,278 edges, 7 classes, collected by Yang et al. (ICML 2016). Each record contains node features, citation edges, class labels. Cora network with bag-of-words features. Suitable for node classification, semi-supervised learning, graph neural networks.
An adolescent friendship network from the Add Health study (2,539 students, 12,969 edges), compiled by James Moody. Each record contains directed edges: True, weighted: True, signed: False. This directed network was created from a survey that took place in 1994/1995. Each student was asked to list his 5 best female and his 5 male friends. A node represents a student and an edge between two students shows that the left student chose the right student as a friend. Higher edge weights ind. Suitable for social network analysis, community detection.
A Human protein interaction network (Figeys lab) dataset of 2,239 nodes and 6,452 edges, collected by D. Figeys et al.. Each record contains directed: True, weighted: False. Human protein-protein interaction network from the Figeys lab.. Suitable for biological network analysis, protein interaction research.
A Blogger.com (2004) dataset of 2,200 bloggers, 2,557 posts, collected by Schler et al. for AAAI 2006. Each record contains blog post text, gender, age, industry, astrological sign. Historical blog corpus from 2004 with rich metadata per blogger (gender, age, industry, astrological sign). Balanced across age groups and genders.. Suitable for authorship attribution, stylistic analysis, demographic language variation, historical NLP.
BlogCatalog
Gnutella
Usenet
Rumble
Twitter (X)
Blogger