A MADOC (Voat, greatawakening) dataset of 11,863 users and 155,890 reply edges from 324,835 interactions, collected by Dankulov et al. via platform APIs, published at ICWSM 2025. Each record contains user-user reply edges, post and comment text, per-record sentiment scores (VADER and TextBlob), subjectivity scores, and toxicity scores (ToxiGen). The dataset is notable for its FAIR-compliant dataset with pre-computed NLP quality labels (sentiment, subjectivity, toxicity) per record, enabling network analysis and content quality research without additional annotation. Intended for research into social network analysis, toxicity detection, cross-platform community dynamics, misinformation research, and hate speech propagation studies.
A MADOC (Voat, milliondollarextreme) dataset of 3,063 users and 10,819 reply edges from 21,190 interactions, collected by Dankulov et al. via platform APIs, published at ICWSM 2025. Each record contains user-user reply edges, post and comment text, per-record sentiment scores (VADER and TextBlob), subjectivity scores, and toxicity scores (ToxiGen). The dataset is notable for its FAIR-compliant dataset with pre-computed NLP quality labels (sentiment, subjectivity, toxicity) per record, enabling network analysis and content quality research without additional annotation. Intended for research into social network analysis, toxicity detection, cross-platform community dynamics, misinformation research, and hate speech propagation studies.
A MADOC (Voat, pics) dataset of 7,712 users and 18,229 reply edges from 38,555 interactions, collected by Dankulov et al. via platform APIs, published at ICWSM 2025. Each record contains user-user reply edges, post and comment text, per-record sentiment scores (VADER and TextBlob), subjectivity scores, and toxicity scores (ToxiGen). The dataset is notable for its FAIR-compliant dataset with pre-computed NLP quality labels (sentiment, subjectivity, toxicity) per record, enabling network analysis and content quality research without additional annotation. Intended for research into social network analysis, toxicity detection, cross-platform community dynamics, misinformation research, and hate speech propagation studies.
A MADOC (Voat, technology) dataset of 14,397 users and 41,519 reply edges from 83,185 interactions, collected by Dankulov et al. via platform APIs, published at ICWSM 2025. Each record contains user-user reply edges, post and comment text, per-record sentiment scores (VADER and TextBlob), subjectivity scores, and toxicity scores (ToxiGen). The dataset is notable for its FAIR-compliant dataset with pre-computed NLP quality labels (sentiment, subjectivity, toxicity) per record, enabling network analysis and content quality research without additional annotation. Intended for research into social network analysis, toxicity detection, cross-platform community dynamics, misinformation research, and hate speech propagation studies.
A MADOC (Voat, videos) dataset of 10,756 users and 31,132 reply edges from 85,197 interactions, collected by Dankulov et al. via platform APIs, published at ICWSM 2025. Each record contains user-user reply edges, post and comment text, per-record sentiment scores (VADER and TextBlob), subjectivity scores, and toxicity scores (ToxiGen). The dataset is notable for its FAIR-compliant dataset with pre-computed NLP quality labels (sentiment, subjectivity, toxicity) per record, enabling network analysis and content quality research without additional annotation. Intended for research into social network analysis, toxicity detection, cross-platform community dynamics, misinformation research, and hate speech propagation studies.
A Manosphere forum (A Voice For Men) dataset of 211,070 posts from 1,802 unique authors on A Voice For Men, collected by Ribeiro et al. via web crawl (2021). Each record contains post text (text_post), author (author), timestamp, post metadata. 211,070 posts from A Voice For Men — part of the Evolution of the Manosphere dataset covering incel, MGTOW, PUA, and related anti-feminist communities across 9 forums. Suitable for manosphere discourse analysis, incel radicalization research, hate speech detection, community evolution, misogynistic content analysis, toxicity measurement.
A Incels.co (manosphere forum) dataset of 2,436,297 posts from 5,933 unique authors across 117,592 threads, collected by Farrell et al. via web crawl (2019-2020). Each record contains post text, author username, author title/flair, join date, post count, post timestamp, thread URL, reply interactions, embedded links. Incels.co dataset contains 2,436,297 forum posts from 2017-11 to 2019-06, part of the iDRAMA manosphere collection covering incel and anti-feminist communities. Suitable for misogyny detection, hate speech research, radicalization analysis, manosphere community studies, extremist forum analysis, NLP classification, social network analysis.
A Manosphere forum (Love-Shy) dataset of 29,232 posts from 826 unique authors on Love-Shy, collected by Ribeiro et al. via web crawl (2021). Each record contains post text (text_post), author (author), timestamp, post metadata. 29,232 posts from Love-Shy — part of the Evolution of the Manosphere dataset covering incel, MGTOW, PUA, and related anti-feminist communities across 9 forums. Suitable for manosphere discourse analysis, incel radicalization research, hate speech detection, community evolution, misogynistic content analysis, toxicity measurement.
A Manosphere forum (Red Pill Talk) dataset of 884,001 posts from 7,900 unique authors on Red Pill Talk, collected by Ribeiro et al. via web crawl (2021). Each record contains post text (text_post), author (author), timestamp, post metadata. 884,001 posts from Red Pill Talk — part of the Evolution of the Manosphere dataset covering incel, MGTOW, PUA, and related anti-feminist communities across 9 forums. Suitable for manosphere discourse analysis, incel radicalization research, hate speech detection, community evolution, misogynistic content analysis, toxicity measurement.
A Manosphere forum (Roosh V Forum) dataset of 1,492,561 posts from 13,599 unique authors on Roosh V Forum, collected by Ribeiro et al. via web crawl (2021). Each record contains post text (text_post), author (author), timestamp, post metadata. 1,492,561 posts from Roosh V Forum — part of the Evolution of the Manosphere dataset covering incel, MGTOW, PUA, and related anti-feminist communities across 9 forums. Suitable for manosphere discourse analysis, incel radicalization research, hate speech detection, community evolution, misogynistic content analysis, toxicity measurement.
A Math Overflow (ALL) dataset of 24,818 users and 239,978 temporal edges, collected by Paranjape, Benson, Leskovec via Stack Exchange Data Dump (2017). Each record contains directed temporal edges (all) between users with Unix timestamps. The dataset is notable for its ALL interaction type from a 6.4-year temporal network of academic Q&A. Intended for research into temporal network analysis, motif discovery, social interaction dynamics, and academic community analysis.
A Math Overflow (A2Q) dataset of 21,688 users and 90,489 temporal edges, collected by Paranjape, Benson, Leskovec via Stack Exchange Data Dump (2017). Each record contains directed temporal edges (a2q) between users with Unix timestamps. The dataset is notable for its A2Q interaction type from a 6.4-year temporal network of academic Q&A. Intended for research into temporal network analysis, motif discovery, social interaction dynamics, and academic community analysis.
A Math Overflow (C2A) dataset of 13,840 users and 81,121 temporal edges, collected by Paranjape, Benson, Leskovec via Stack Exchange Data Dump (2017). Each record contains directed temporal edges (c2a) between users with Unix timestamps. The dataset is notable for its C2A interaction type from a 6.4-year temporal network of academic Q&A. Intended for research into temporal network analysis, motif discovery, social interaction dynamics, and academic community analysis.
A Math Overflow (C2Q) dataset of 16,836 users and 101,329 temporal edges, collected by Paranjape, Benson, Leskovec via Stack Exchange Data Dump (2017). Each record contains directed temporal edges (c2q) between users with Unix timestamps. The dataset is notable for its C2Q interaction type from a 6.4-year temporal network of academic Q&A. Intended for research into temporal network analysis, motif discovery, social interaction dynamics, and academic community analysis.
A Meme Descriptions (Cheezburger dataset) dataset of 252,284 meme descriptions and 117 images, collected by Wang and Wen via Cheezburger (2015). Each record contains meme image and text descriptions, popularity scores. The dataset is notable for its multimodal dataset for meme popularity prediction and generation. Intended for research into multimodal learning, meme generation, popularity prediction, and computational humor.
A Web (news sites and blogs, Aug 2008–Apr 2009) dataset of 71,568 phrase clusters, 310,457 unique phrase nodes, 238,889 directed root→variant edges, 71,568 phrases in text corpus, collected by Leskovec et al. via web crawl of 1.5M articles/day, August 2008 to April 2009, distributed by SNAP. Each record contains phrase cluster graph where nodes are natural language phrases extracted from news articles and blogs; directed edges from root phrase to variant phrases within each cluster; edge weight = variant phrase frequency; covers the 2008 US election, financial crisis, and other major news events. The dataset is notable for its foundational information diffusion dataset capturing how phrases mutate and propagate across mainstream media and blogs; the cluster graph encodes phrase similarity and co-occurrence structure, enabling study of linguistic variation in news cycles. Intended for research into information diffusion modeling, meme propagation analysis, phrase clustering, linguistic variation in news media, temporal network analysis, and graph-based NLP research.
A Meneame dataset of 14,995 articles, 20,000 comments, 45,392 users, collected by Meneame Media Bias Team. Each record contains article title, text, bias labels, sentiment, emotion scores, karma features, user-outlet graph. Spanish social news aggregator dataset with media bias labels and rich interaction features. Suitable for media bias detection, sentiment analysis, social news dynamics, graph analysis.
A twitter dataset of 957,956 tweet IDs from the #MeToo movement (Q3 2020), collected by Maiorana, Morales Henry, and Weintraub via Social Feed Manager. Each record contains tweet IDs only — full content requires hydration. Quarterly snapshot of #MeToo discourse enabling longitudinal analysis when combined with other quarterly collections. Suitable for social movement analysis, gender studies, harassment detection, discourse trends.
A News Articles (multiple sources) dataset of 85,917 articles (43,610 fake, 34,941 true, 7,366 russian propaganda), collected by Steven Peutz via Kaggle (2022); Russian propaganda subset sourced from the EUvsDisinfo Project (2015-present). Each record contains full article text and veracity label (fake, true, or russian_propaganda). combination of general fake/true news articles with a dedicated Russian propaganda subset from the EUvsDisinfo database, enabling comparative analysis across disinformation types. Suitable for fake news detection, misinformation topic modeling, propaganda characterization, and classification benchmarking.
A MIT Media Lab (mobile phone Bluetooth proximity sensing) dataset of 96 nodes and 2,539 edges, collected by Nathan Eagle & Alex Pentland via mobile phone study. Each record contains directed: False, weighted: True, signed: False, temporal: True. Bluetooth proximity contact network of 100 MIT students and faculty over 9 months. One of the earliest large-scale mobile sensing datasets, with temporal edges capturing when people were physically near each other.. Suitable for temporal network analysis, proximity sensing, community detection, mobility.
Moltbook, a social network whose members are autonomous AI agents (Hugging Face jscmp4/Moltbook, 2025): 100,590 agent profiles, 3,352,520 posts and 16,663,266 comments. The graph links agents to the posts and comments they wrote and comments to their posts (20,199,253 nodes, 36,678,988 edges). Suitable for research on AI agent behaviour and social dynamics.
A twitter dataset of 255,363 tweet IDs about the 2022 monkeypox outbreak, collected by Thakur (University of Cincinnati, 2022). Each record contains tweet IDs only — full content requires hydration. First public Twitter dataset on the 2022 monkeypox outbreak, enabling early-stage disease discourse analysis. Suitable for disease surveillance, public health communication, health misinformation.
A Windsurfer social interaction at a beach dataset of 43 nodes and 336 edges, collected by Freeman et al.. Each record contains directed edges: False, weighted: True, signed: False. This undirected network contains interpersonal contacts between windsurfers in southern California during the fall of 1986. A node represents a windsurfer and an edge between two windsurfers shows that there was a interpersonal contact.. Suitable for social network analysis, community detection.
A Bison dominance interaction observation dataset of 26 nodes and 314 edges, collected by KONECT. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains dominance between American bisons in 1972 on the National Bison Range in Moiese (Montana). A node represents a bison and an edge represents dominance of the left bison over the right bison. The edge values indicate how often this dominance behaviour was observed.. Suitable for animal social network analysis.
A Political blogosphere (US 2004 election) dataset of 1,224 nodes and 19,025 edges, collected by Adamic & Glance via blog crawl. Each record contains directed edges: True, weighted: False, signed: False. This directed network contains front-page hyperlinks between blogs in the context of the 2004 US election. A node represents a blog and an edge represents a hyperlink between two blogs.. Suitable for political polarization, community detection, information diffusion.
A Cattle dominance interaction observation dataset of 28 nodes and 217 edges, collected by KONECT. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains dominance behaviours observed between dairy cattles at the Iberia Livestock Experiment Station in Jenerette, Louisiana. A node represents a cattle and an edge represents dominance of the left cattle over the right cattle. The edge values indicate how often this dominan. Suitable for animal social network analysis.
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 Hen dominance pecking order observation dataset of 32 nodes and 496 edges, collected by KONECT. Each record contains directed edges: True, weighted: False, signed: False. This directed network contains the peck order of 32 White Leghorn hens observed in 1946. A node represents a hen and an edge represents dominance of the left hen over the right hen.. Suitable for animal social network analysis, hierarchy detection.
A Small highschool in Illinois (offline survey, 1957–1958) dataset of 70 nodes and 366 edges, collected by J.L. Moreno via sociometric survey. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains friendships between boys in a small highschool in Illinois. Each boy was asked once in the fall of 1957 and the spring of 1958. This dataset aggregates the results from both dates. A node represents a boy and an edge between two boys shows that the left boy chose the r. Suitable for community detection, friendship network analysis, social influence.
A Medical innovation adoption network dataset of 241 nodes and 1,098 edges, collected by Coleman et al.. Each record contains directed edges: True, weighted: False, signed: False. This directed network captures innovation spread among 246 physicians in for towns in Illinois, Peoria, Bloomington, Quincy and Galesburg. The data was collected in 1966. A node represents a physician and an edge between two physicians shows that the left physician told that the righ physician is hi. Suitable for diffusion of innovations, social influence.
A Kangaroo dominance interaction observation dataset of 17 nodes and 91 edges, collected by KONECT. Each record contains directed edges: False, weighted: True, signed: False. This undirected network contains interactions between free-ranging grey kangaroos. A node represents a kangaroo and an edge between two kangaroos shows that there was an interaction between them. The edge values denote the total count of interactions.. Suitable for animal social network analysis, dominance hierarchies.
A Literary character co-occurrence (Les Misérables) dataset of 77 nodes and 254 edges, collected by D.E. Knuth. Each record contains directed edges: False, weighted: True, signed: False. This undirected network contains co-occurances of characters in Victor Hugo's novel 'Les Misérables'. A node represents a character and an edge between two nodes shows that these two characters appeared in the same chapter of the the book. The weight of each link indicates how often such a co-appear. Suitable for community detection benchmark, graph algorithm testing.
A Macaque grooming behavior observation dataset of 62 nodes and 1,187 edges, collected by KONECT. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains dominance behaviour in a colony of 62 adult female Japanese macaques (<i>Macaca fuscata fuscata</i>). A node represents a macaque and a directed edge A → B represents dominance of macaque A over macaque B.. Suitable for animal social network analysis.
A Name association network dataset of 1,773 nodes and 9,131 edges, collected by J.L. Moreno. Each record contains directed edges: False, weighted: True, signed: False. This undirected network contains nouns (places and names) of the King James Version of the Bible and information about their co-occurrences. A node represents one of the above noun types and an edge indicates that two nouns appeared together in the same Bible verse. The edge multiplicity denotes ho. Suitable for semantic network analysis.
A friendship-rating network of 217 residents of a residence hall at the Australian National University (217 nodes, 2,672 edges), collected by Freeman, Webster & Kirke. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains friendship ratings between 217 residents living at a residence hall located on the Australian National University campus. A node represents a person and. Suitable for social network analysis, community detection.
A Protein-protein interaction network dataset of 1,870 nodes and 2,277 edges, collected by KONECT. Each record contains directed edges: False, weighted: False, signed: False. This undirected network contains protein interactions contained in yeast. Research showed that proteins with a high degree were more important for the surivial of the yeast than others. A node represents a protein and an edge represents a metabolic interaction between two proteins. The network cont. Suitable for biological network analysis, community detection.
A Rhesus monkey grooming observation dataset of 16 nodes and 111 edges, collected by KONECT. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains observed grooming episodes between free ranging rhesus macaques (<i>Macaca mulatta</i>) in Cayo Santiago during a two month period in 1963. Cayo Santiago is an island off the coast of Puerto Rico, also known as Isla de los monos (Island of the monkeys). A node repres. Suitable for animal social network analysis.
A Monastery interpersonal relations survey dataset of 18 nodes and 189 edges, collected by S.F. Sampson. Each record contains directed edges: True, weighted: True, signed: True. This directed network contains ratings between monks related to a crisis in a cloister (or monastery) in New English (USA) which lead to a departure of several of the monks. This dataset aggregates several available ratings ((dis)esteem, (dis)liking, positive/negative influence, praise/blame) into o. Suitable for signed network analysis, community detection, balance theory.
A Seventh grade classroom friendship survey dataset of 29 nodes and 376 edges, collected by J.L. Moreno. Each record contains directed edges: True, weighted: True, signed: False. This directed network contains proximity ratings between studetns from 29 seventh grade students from a school in Victoria. Among other questions the students were asked to nominate their preferred classmates for three different activities. A node represents a student. An edge between two nodes show. Suitable for social network analysis, community detection.
A Sheep dominance interaction observation dataset of 28 nodes and 250 edges, collected by KONECT. Each record contains directed edges: True, weighted: True, signed: False. This directed network deescribes the dominance relationships between 28 female bighorn sheep (<i>Ovis canadensis</i>) observed on the National Bison Range in western Montana (USA) in 1984. A node represents a female bighorn sheep and an edge represents dominance of the left female bighorn sheep over. Suitable for animal social network analysis.
A Taro exchange network in Papua New Guinea village dataset of 22 nodes and 78 edges, collected by Hage & Harary. Each record contains directed edges: True, weighted: False, signed: False. This undirected network contains gift-givings (taro) between households in a Papuan village. A node represents a household and an edge between two households indicates that there happened a gift-giving.. Suitable for exchange network analysis, community detection.
A Madrid train bombing terrorist network (2004) dataset of 64 nodes and 243 edges, collected by KONECT. Each record contains directed edges: False, weighted: True, signed: False. This undirected network contains contacts between suspected terrorists involved in the train bombing of Madrid on March 11, 2004 as reconstructed from newspapers. A node represents a terrorist and an edge between two terrorists shows that there was a contact between the two terroists. The edge weigh. Suitable for dark network analysis, community detection.
A University freshman friendship survey (longitudinal) dataset of 32 nodes and 710 edges, collected by Van de Bunt et al.. Each record contains directed edges: True, weighted: True, signed: True. This directed network contains friendship ratings between 32 university freshmen who mostly did not know each other before starting university. Each student was asked to rate the other student at seven different time points. Note that the origin of the timestamps is not accurately known but the dis. Suitable for longitudinal social network analysis, friendship formation. The file holds 3,062 ratings, one row per rating in each survey wave; the edge count of 710 counts distinct pairs.
A Zebra social interaction observation dataset of 27 nodes and 111 edges, collected by KONECT. Each record contains directed edges: False, weighted: False, signed: False. This undirected network contains interactions between 28 Grévy's zebras (<i>Equus grevyi</i>) in Kenya. A node represents a zebra and an edge between two zebras shows that there was an interaction between them during the study.. Suitable for animal social network analysis.
A Twitter (user-hashtag interactions) dataset of 705,632 nodes and 1,890,661 edges, collected by Munmun De Choudhury et al. via Twitter API. Each record contains directed: False, weighted: True, signed: False, temporal: True. Bipartite temporal network linking Twitter users to hashtags they used, with Unix timestamps. Useful for studying hashtag adoption, information diffusion, and user interest modeling.. Suitable for hashtag analysis, bipartite network analysis, information diffusion.
A Twitter (following network sample) dataset of 465,017 nodes and 834,797 edges, collected by Munmun De Choudhury et al. via Twitter API. Each record contains directed: True, weighted: False, signed: False, temporal: False. Large directed Twitter following network of 465k users and 835k edges, collected via API sampling. Unweighted directed graph suitable for large-scale social network analysis.. Suitable for Twitter network analysis, community detection, influence propagation.
A Twitch.tv (live streaming platform) dataset of 34,118 users and 214,556 undirected friendship edges across 6 language communities (DE, ENGB, ES, FR, PTBR, RU) merged into one cumulative graph, collected by Rozemberczki et al. via Twitch API, distributed via MUSAE/SNAP. Each record contains undirected friendship edges between Twitch streamers within each language community; node IDs namespaced by language to avoid collision across subgraphs. The dataset is notable for its multilingual streaming platform social network covering 6 distinct language communities, enabling cross-community structural comparison and multilingual graph embedding research. Intended for research into community detection, graph embedding, node classification, cross-platform social network analysis, and multilingual network comparison.
A MySpace dataset of 854,498 users, 6,489,736 directed follower edges, collected by Zhang and Tang via AMiner. Each record contains directed follower edges (source -> target means source follows target). Large-scale MySpace follower graph from AMiner, capturing social connections among 854K users. Suitable for social network analysis, community detection, influence propagation, graph embedding.
A MySpace dataset of 80,749 messages between 36,459 users (41,533 directed user pairs) over two months, collected by Šuvakov et al. (Dryad, 2013). Each record holds the sender, the recipient, the time in minutes since the start of the window, and the message's SentiStrength positive (1 to 5) and negative (-1 to -5) scores. Suitable for sentiment dynamics, dialogue patterns and temporal network analysis.
A Twitter (Nigerian language communities) dataset of 60,823 tweet IDs across 4 Nigerian languages with sentiment labels, collected by Muhammad et al. via Twitter API (2022). Each record contains tweet ID, sentiment label (positive/negative/neutral), language, split (train/dev/test). First large-scale sentiment corpus for Nigerian languages — 60,823 annotated tweet IDs across Hausa, Igbo, Yoruba, and Nigerian Pidgin, enabling low-resource African NLP research.. Suitable for low-resource NLP, African language sentiment analysis, multilingual classification, cross-lingual transfer..
Voat
Incels.co
Stack Exchange
Menéame
Twitter (X)
Moltbook
Twitch
Myspace