A Twitter dataset of 5,072,124 tweets from 1,835 congressional accounts, collected by Alex Litel via Twitter API (2017-2023). Each record contains tweet text, screen name, user ID, timestamp, retweet count, favorite count, source, retweet status. A daily collection of tweets from U.S. Congress members, committees, caucuses and party accounts, one JSON file per day; 11 days have no file in the source repository (for example 2022-01-15, 2022-03-17 and 2022-10-24 to 2022-10-31). metadata/historical-users-filtered.json, from the collection's companion repository, maps each account to its member or office, chamber, party and type. Suitable for political communication, sentiment analysis, election studies, congressional behavior, social media analysis.
A Twitch (gaming social network) dataset of 168,116 users and 6,797,558 mutual follower edges, collected by Rozemberczki & Sarkar via Twitch public API (2018). Each record contains undirected mutual follower edges, node features (views, mature, life_time, created_at, updated_at, dead_account, language, affiliate). The dataset is notable for its large-scale strongly connected component with no missing attributes, designed for node classification and count data regression. Intended for research into social network analysis, node classification, community detection, and transfer learning in gaming communities.
A NDS-VU signed network repository dataset of 138,588 Wikipedia editors and 715,883 signed interaction edges on political articles, collected by NDS-VU from Wikipedia edit history. Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Large signed network of Wikipedia editors working on political articles, capturing support and opposition relationships derived from edit patterns and talk page interactions in politically charged content areas. Suitable for political discourse analysis, signed network analysis, editor behavior.
A NDS-VU signed network repository dataset of 7,116 Wikipedia editors and 1,048,576 signed conflict/collaboration edges, collected by NDS-VU from Wikipedia edit history. Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Large signed network of Wikipedia editor interactions derived from edit conflict and revert patterns, where positive edges indicate collaborative edits and negative edges indicate conflicts and reverts between editors. Suitable for signed network analysis, conflict detection, collaborative platform dynamics.
A Twitter (hateful user network) dataset of 100,386 users (544 hateful, 4,427 normal, 0 suspended) with 2,286,592 follower edges, collected by Ribeiro et al. via Twitter API (2018). Each record contains user ID, hate label (hateful/normal/suspended), follower/followee/statuses counts, network centrality metrics, 200+ Empath emotional category scores, 300-dim GloVe embeddings, sentiment, subjectivity, tweet engagement statistics. One of the most feature-rich hateful user datasets, with 100,386 Twitter users labeled for hatefulness and supplemented with full network topology, Empath psychological features, and GloVe embeddings. Suitable for hate speech detection, hateful user classification, network-based toxicity analysis, suspended account detection, social influence of hate, graph-based hate propagation.
A Twitter (crowdsourced hate speech annotation) dataset of 24,783 tweets labeled as hate speech (1,430), offensive (19,190), or neither (4,163), collected by Davidson et al. via CrowdFlower crowdsourcing (2017). Each record contains tweet text, majority class label (0=hate/1=offensive/2=neither), per-label coder counts, total coder count. Davidson et al. 2017 is one of the most cited hate speech datasets, with 24,783 tweets annotated by multiple crowdworkers — distinguished by separating hate speech from merely offensive language. Suitable for hate speech detection, offensive language classification, content moderation, NLP benchmarking, crowdsourced annotation research.
23,992 YouTube comments with sentiment and tone labels and short pros/cons notes, published on Hugging Face by AnandforU, in train (19,992), validation (2,000) and test (2,000) splits. Each record holds the comment text, sentiment, tone, pros and cons. Suitable for sentiment analysis and comment moderation.
A Twitter dataset of 11,929 labeled financial tweets, collected by zeroshot via Twitter API. Each record contains tweet text, sentiment label (positive/negative/neutral), sentiment value (0/1/2), train/valid split. A labeled dataset of tweets about financial news, with sentiment annotations for stock market-related content. Suitable for sentiment analysis, financial NLP, stock market prediction, social media analytics, natural language processing.
A NDS-VU signed network repository dataset of 10,885 users and 251,406 signed interaction edges from Italian referendum Twitter, collected by NDS-VU (Vrije Universiteit Amsterdam, 2016). Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Signed retweet/mention network from the 2016 Italian constitutional referendum on Twitter, where edges represent support (+1) or opposition (-1) interactions between users discussing the referendum. Suitable for political polarization, signed social network analysis, referendum discourse.
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 NDS-VU signed network repository dataset of 7,116 Wikipedia users and 100,693 signed RfA vote edges, collected by NDS-VU from Wikipedia RfA archives. Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Signed network of Wikipedia requests for adminship (RfA) votes where edges represent support (+1) or oppose (-1) votes cast during admin election campaigns, enabling trust and reputation analysis in collaborative platforms. Suitable for signed network analysis, reputation systems, voting behavior, trust modeling.
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 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 Twitter dataset of 4,465 labeled Twitter accounts (991 bots, 3474 genuine), collected by Ashu273k — curated subset of Cresci et al. (2017). Each record contains user profile features (followers, friends, statuses, favourites, listed count), binary bot/genuine label, user bio text. Small, clean subset of the Cresci-2017 dataset curated for rapid prototyping and benchmarking of bot detection models, with pre-extracted user-level features and binary labels. Suitable for bot detection, social spambot classification, feature selection for account authenticity, rapid prototyping of detection algorithms.
A NDS-VU signed network repository dataset of 220 Congress members and 521 signed co-sponsorship edges, 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. Signed network of U.S. Congress members based on bill co-sponsorship patterns, where positive edges indicate co-sponsorship and negative edges indicate opposition, enabling political alignment and polarization analysis. Suitable for political polarization, signed network analysis, legislative behavior.
A Multiple (Wikipedia, GDELT, Google Trends) dataset of 51 weeks of cross-platform attention data across 6 regions, collected by Luke Steuber via Wikipedia API, GDELT Project, and Google Trends (2020-2025). Each record contains weekly attention scores by region, component metrics (Google Trends, GDELT, Wikipedia), and regional composite scores. A unified, weekly time-series of public attention to the United States from three independent sources, broken down by global region, enabling multi-perspective analysis of attention dynamics. Suitable for information cascade, event detection, public health monitoring, political communication, media analysis, attention tracking, regional sentiment analysis.
A NDS-VU signed network repository dataset of 19 nodes and 125 signed edges from a monastery social network, collected by Sampson (1969), distributed by NDS-VU. Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Classic small signed social network from a monastery study by Sampson (1969), widely used as a benchmark for signed community detection and balance theory validation. Suitable for signed network analysis, balance theory, community detection benchmarking.
A NDS-VU signed network repository dataset of 16 tribes and 58 signed alliance/opposition edges, collected by Read (1954), distributed by NDS-VU. Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Classic ethnographic signed network of New Guinea Highland tribes with alliance (+1) and opposition (-1) relationships, one of the earliest and most cited networks for structural balance theory validation. Suitable for balance theory, structural balance, signed network benchmarking.
Twitter (X)
Twitch
Wikipedia
YouTube
Google