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 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..
A NELA-PS (pink slime local news) dataset of 7,910,965 articles from 1,093 pink slime news sources, collected by Horne and Gruppi, published via Harvard Dataverse (2024). Each record contains full article text, headline, source outlet name, publication date, author, location, network, and URL. The dataset is notable for its largest available dataset of pink slime news — low-quality partisan content systematically framed as authentic local news, spanning 1,093 outlets over 2.5 years with stratified cross-outlet topic modeling. Intended for research into misinformation detection, local news ecosystem analysis, partisan content identification, and pink slime outlet characterisation.
A Twitter dataset of 404,615 nodes, 434,963 edges, collected by Pierri et al. (OSoMe/Indiana University + Politecnico di Milano), 2020. Each record contains retweet network (uid1, uid2), user-domain bipartite network, user co-sharing network. Twitter news-sharing network revealing information polluters through retweet and co-sharing patterns.. Suitable for Information polluter identification, news sharing network analysis, source credibility..
A Reddit dataset of 398,487 comments from NFL subreddits during the 2020 draft, collected by Caio Brighenti via Reddit API (PRAW). Each record contains comment body, team, timestamp, VADER sentiment score, author. A real-time snapshot of NFL fan sentiment during the 2020 draft, capturing reactions to each pick across team subreddits. Suitable for sentiment analysis, sports analytics, social media analysis, fan behavior, natural language processing.
A Twitter dataset of 5,171 sentiment-labeled tweets about NFL teams, collected by Cameron C (Kaggle user cammyc) via Twitter API. Each record contains tweet text, sentiment (positive/negative/neutral), team, username, timestamp, confidence score. A curated dataset of 3,701 NFL-related tweets with sentiment labels, enabling research on sports fan sentiment, team-specific public opinion, and social media engagement patterns during the NFL season. Suitable for sentiment analysis, sports marketing, fan engagement, public opinion tracking, temporal sentiment analysis.
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 Twitter (aggregate) dataset of 4,709,404 topic-related tweets across 6 social issues, collected by NORC at the University of Chicago. Each record contains date, topic, daily volume, VADER compound sentiment, pos/neg ratio, account type (community/bot/news/government/activist), US state. Aggregate Twitter data on six politically salient social issues with daily volume and VADER sentiment scores 2016-2022, including unique bot-vs-community account type breakdowns and US state-level marijuana discourse data. Suitable for political communication, sentiment trends, bot detection, state-level public opinion, longitudinal social issue analysis.
A OECD dataset of 44 countries spanning 2008–2023, collected by OECD via Zenodo. Each record contains Country, year, indicator, observation value, breakdowns by age/gender/education. Detailed ICT usage data with demographic breakdowns. Suitable for digital divide analysis, ICT adoption trends.
A Twitter (OffensEval 2020 — 5 languages) dataset of 61,922 tweets across 5 languages with offensive/non-offensive labels, collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), language, split. Combined OffensEval 2020 benchmark across Arabic, Danish, English, Greek, and Turkish — 61,922 tweets from the most-participated SemEval task ever (528 teams). Suitable for multilingual offensive language detection, cross-lingual transfer, content moderation, NLP benchmarking.
A Twitter (OffensEval 2020 — Arabic) dataset of 9,666 Arabic tweets (1,919 offensive, 7,747 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 Arabic — 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 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 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 Twitter (OffensEval 2020 — Greek) dataset of 10,287 Greek tweets (2,728 offensive, 7,559 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 Greek — 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 Twitter (OffensEval 2020 — Turkish) dataset of 34,792 Turkish tweets (6,757 offensive, 28,035 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 Turkish — 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 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 Academic co-authorship bipartite network dataset of 38,741 nodes and 58,595 edges, collected by T. Opsahl. Each record contains directed: False, weighted: False, signed: False. Large bipartite collaboration network of authors and papers — useful for studying academic collaboration patterns.. Suitable for bipartite network analysis, academic collaboration, community detection.
A Southern women event attendance bipartite network (1930s) dataset of 32 nodes and 89 edges, collected by T. Opsahl. Each record contains directed: False, weighted: False, signed: False. Bipartite network of women and social events — Opsahl's version of the classic Davis dataset.. Suitable for bipartite network analysis, classic network benchmark.
A UC Irvine online student forum dataset of 1,421 nodes and 7,089 edges, collected by T. Opsahl. Each record contains directed: False, weighted: True, signed: False. Temporal bipartite network of students posting to forum threads with Unix timestamps.. Suitable for temporal bipartite network analysis, online community analysis.
A UC Irvine online messaging system dataset of 1,899 nodes and 20,296 edges, collected by T. Opsahl. Each record contains directed: True, weighted: True, signed: False. Temporal directed messaging network among UC Irvine students with Unix timestamps.. Suitable for temporal network analysis, online social network analysis. The file holds 59,835 messages, one row per message; the edge count of 20,296 counts distinct pairs.
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.
Orkut users and the groups they joined, as distributed by KONECT (orkut-groupmemberships), from the crawl of Mislove et al. (2007): a bipartite network linking 2.8 million users to the 8.7 million Orkut groups they belonged to, with some 327 million memberships. Orkut was Google's social network, active 2004–2014. Useful for large-scale community detection, affiliation-network analysis and the structure of online groups.
The Orkut friendship network crawled by Mislove et al. (2007), as distributed by Network Repository (soc-orkut-dir): 3,072,441 users and 117,185,083 undirected friendships, each pair listed once. Despite the "-dir" in Network Repository's name, the file is marked symmetric and Orkut friendships were mutual. Orkut was Google's social network, active 2004–2014 and especially popular in Brazil and India. Useful for large-scale social network analysis, community detection and graph mining.
A Parler (alternative social network) dataset of ~183,000,000 posts from Parler's full history before deplatforming, collected by Aliapoulios et al. via Parler API (2021). Each record contains post body text, username, timestamp, hashtags, upvotes, reposts, impressions, comment count, sensitive flag, media attachments, reply depth, parent post ID. Complete Parler post archive from before the platform's deplatforming — includes Jan 6 Capitol riot content, far-right discourse, and cross-posts from Twitter. Suitable for deplatforming research, far-right discourse analysis, January 6th research, hate speech detection, alternative platform behavior, content moderation effects.
A Parler (alternative social network) dataset of 13,254,060 Parler user profiles with bios and activity metrics, collected by Aliapoulios et al. via Parler API (2021). Each record contains user ID, username, bio text, follower/following counts, post/comment/like/media counts, join date, verified/banned/human flags. 13,254,060 Parler user profiles from the platform's peak period before its deplatforming — 605 verified accounts, 287,391 banned users. Suitable for deplatforming research, far-right social network analysis, user behavior on alternative platforms, bot detection, political communication research.
A Petster.com (pet social network — carnivores) dataset of 623,766 nodes and 15,699,276 edges, collected by KONECT via Petster crawl. Each record contains directed: False, weighted: False, signed: False. Friendship network of carnivore pet profiles on a pet social networking site.. Suitable for social network analysis, community detection.
A Petster.com (pet social network — cats) dataset of 149,700 nodes and 5,449,275 edges, collected by KONECT via Petster crawl. Each record contains directed: False, weighted: False, signed: False. Friendship network of cat profiles on a pet social networking site.. Suitable for social network analysis, community detection.
A Petster.com (pet social network — dogs) dataset of 426,820 nodes and 8,546,581 edges, collected by KONECT via Petster crawl. Each record contains directed: False, weighted: False, signed: False. Friendship network of dog profiles on a pet social networking site.. Suitable for social network analysis, community detection.
A Petster.com (pet social network — hamsters) dataset of 1,858 nodes and 12,534 edges, collected by KONECT via Petster crawl. Each record contains directed: False, weighted: False, signed: False. Friendship network of hamster profiles on a pet social networking site.. Suitable for social network analysis, community detection.
A Pew Research Center dataset of 11 annual survey waves spanning 2012–2025, collected by Pew Research Center via annual US adult surveys. Each record contains survey year, platform-specific adoption percentages for US adults. Annual survey tracking US adult adoption rates across all major social media platforms from 2012 to the present. Suitable for social media adoption trends, platform popularity analysis, digital behavior research, longitudinal survey analysis.
Twitter conversations around the Charlie Hebdo shooting in Paris (January 2015), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 2,079 threads, each a source tweet and its replies, of which 458 were judged rumours by journalists and 1,621 not. Every rumour also carries its veracity as later established: 193 true, 116 false and 149 unverified. 38,268 tweets in all, with full tweet JSON. ECHO builds the reply network (36,189 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
Twitter conversations around the Ferguson unrest after the shooting of Michael Brown (August 2014), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 1,143 threads, each a source tweet and its replies, of which 284 were judged rumours by journalists and 859 not. Every rumour also carries its veracity as later established: 10 true, 8 false and 266 unverified. 24,175 tweets in all, with full tweet JSON. ECHO builds the reply network (22,260 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
Twitter conversations around the Germanwings plane crash in the French Alps (March 2015), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 469 threads, each a source tweet and its replies, of which 238 were judged rumours by journalists and 231 not. Every rumour also carries its veracity as later established: 94 true, 111 false and 33 unverified. 4,489 tweets in all, with full tweet JSON. ECHO builds the reply network (4,020 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
Twitter conversations around the Ottawa Parliament Hill shooting (October 2014), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 890 threads, each a source tweet and its replies, of which 470 were judged rumours by journalists and 420 not. Every rumour also carries its veracity as later established: 329 true, 72 false and 69 unverified. 12,284 tweets in all, with full tweet JSON. ECHO builds the reply network (11,394 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
Twitter conversations around the Sydney Lindt café siege (December 2014), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 1,221 threads, each a source tweet and its replies, of which 522 were judged rumours by journalists and 699 not. Every rumour also carries its veracity as later established: 382 true, 86 false and 54 unverified. 23,996 tweets in all, with full tweet JSON. ECHO builds the reply network (22,775 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
A Pokec dataset of 1,632,803 users, 22,301,964 directed relationships, collected by Takac and Zabovsky via Pokec platform (2012). Each record contains directed edges between Pokec users representing user-to-user relationships in the Slovak social network. Pokec social network, the most popular social network in Slovakia, capturing user relationships in a regional online community. Suitable for social network analysis, community detection, graph mining, regional social network studies, link prediction.
A Twitter dataset of 128,484 tweets with political leaning labels, collected by Chatelet via Hugging Face. Each record contains text, political label (liberal, conservative, neutral). 100K tweets labeled for political ideology with balanced classes. Suitable for political NLP, ideology classification, discourse analysis.
A Twitter dataset of 18470 nodes, 48365 edges, collected by Nikolov et al. (OSoMe/Indiana University), 2010. Each record contains retweet edgelist, mention edgelist, node political labels (left/right/-). Twitter retweet and mention networks from 2010 with political leaning labels per user; three graph types for polarization analysis.. Suitable for Political polarization, echo chamber detection, social bubble measurement..
A Prosper.com (peer-to-peer lending platform) dataset of 89,269 nodes and 3,330,225 edges, collected by KONECT via Prosper API. Each record contains directed: True, weighted: True, signed: False, temporal: True. Directed trust network from a peer-to-peer lending platform where users vouch for borrowers. Temporal edges with Unix timestamps enable longitudinal trust propagation analysis.. Suitable for trust network analysis, temporal network analysis, peer-to-peer lending.
Twitch
Myspace
Twitter (X)
Reddit
Facebook
Instagram
Orkut
Parler
Petster
Pokec