The wikilink network of the German Wikipedia, as distributed by KONECT (wikipedia_link_de), built from the Wikimedia dumps: 3,225,565 articles and 81,626,917 directed links between them, where a link from A to B means article A links to article B (article namespace only; 31,722 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 3,603,726 articles and 96,865,851 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the French Wikipedia, as distributed by KONECT (wikipedia_link_fr), built from the Wikimedia dumps: 3,023,165 articles and 102,382,410 directed links between them, where a link from A to B means article A links to article B (article namespace only; 17,100 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 3,333,397 articles and 123,709,902 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the Italian Wikipedia, as distributed by KONECT (wikipedia_link_it), built from the Wikimedia dumps: 1,865,965 articles and 91,555,008 directed links between them, where a link from A to B means article A links to article B (article namespace only; 27,438 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 2,148,791 articles and 104,719,994 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the Japanese Wikipedia, as distributed by KONECT (wikipedia_link_ja), built from the Wikimedia dumps: 1,610,638 articles and 71,055,717 directed links between them, where a link from A to B means article A links to article B (article namespace only; 6,153 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 1,767,268 articles and 83,202,622 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the Polish Wikipedia, as distributed by KONECT (wikipedia_link_pl), built from the Wikimedia dumps: 1,529,135 articles and 57,489,447 directed links between them, where a link from A to B means article A links to article B (article namespace only; 2,211 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 1,688,608 articles and 70,256,468 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the Portuguese Wikipedia, as distributed by KONECT (wikipedia_link_pt), built from the Wikimedia dumps: 1,603,222 articles and 49,021,409 directed links between them, where a link from A to B means article A links to article B (article namespace only; 7,445 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 1,744,781 articles and 57,679,728 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
The wikilink network of the Russian Wikipedia, as distributed by KONECT (wikipedia_link_ru), built from the Wikimedia dumps: 2,853,118 articles and 82,056,101 directed links between them, where a link from A to B means article A links to article B (article namespace only; 20,493 pages link to themselves). Counts are those of ECHO's copy of the KONECT archive; KONECT's current release lists 3,370,462 articles and 93,373,056 links, including articles without links. One of eight language editions on ECHO (English, German, French, Italian, Japanese, Polish, Portuguese, Russian). Useful for knowledge-graph and hyperlink-structure analysis, link prediction, community detection and comparison across language editions.
Posts from 329 thematically distinct 'general' thread collections on 4chan's /pol/ board, November 2013 to May 2019, released by Emilija Jokubauskaitė and Stijn Peeters (Zenodo 3603292). 13,756,579 posts in 58,687 threads, one CSV per general plus an index file. Each row holds the thread ID, post ID, timestamp, body text, subject, poster name, image file name and MD5, and the poster's country code and name. Suitable for studying recurring political discussion threads, propaganda, hate speech and online subcultures.
A 4chan /pol/ (Politically Incorrect board) dataset of 134,529,233 posts across 3,397,911 threads (4,894,179 with 7-score Perspective API annotations), collected by Papasavva et al. via 4chan API + Perspective API (2020). Each record contains post text (com), thread ID, post number, country, timestamp, image metadata, 7 Perspective API scores (TOXICITY, SEVERE_TOXICITY, INFLAMMATORY, PROFANITY, INSULT, OBSCENE, SPAM), named entities. 134,529,233 4chan /pol/ posts with per-post Perspective API toxicity scores across 7 dimensions — one of the most comprehensively annotated fringe platform datasets covering 3.5 years of /pol/ discourse. Suitable for toxicity measurement, hate speech analysis, fringe platform research, political discourse on imageboards, cross-platform toxicity comparison, longitudinal hate speech trends.
A 4chan (/pol/) dataset of 4,998,945 posts from /pol/, collected by kjj0 via 4chan API. Each record contains post text content. A 5 million post sample from 4chan's /pol/ board, capturing political discourse, extremism, and online communities. Suitable for misinformation, propaganda, hate speech, political communication, online communities, extremist discourse.
A 4chan (/pol/) dataset of 499,723 posts with moderation scores, collected by kjj0 via 4chan API and OpenAI moderation model. Each record contains post text, sexual score, hate score, violence score, self-harm score, sexual/minors score, hate/threatening score, violence/graphic score. A dataset of 500,000 /pol/ posts with OpenAI moderation scores, enabling research on content moderation, hate speech detection, and harmful content analysis. Suitable for content moderation, hate speech detection, toxicity analysis, misinformation, political communication.
A Academia.edu dataset of 200,169 academics, 1,022,883 connections, collected by Fire et al. (2011) via Academia.edu platform. Each record contains undirected edges between academics, representing follows or connections between researchers on the platform. Academic social network capturing researcher connections on Academia.edu, a platform for sharing research papers and tracking scholarly impact. Suitable for link prediction, academic network analysis, scholarly collaboration studies, research impact 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 David Copperfield novel (word adjacency network) dataset of 112 nodes and 425 edges, collected by M.E.J. Newman. Each record contains directed: False, weighted: False, signed: False. Word adjacency network of adjectives and nouns in David Copperfield — classic linguistic network benchmark.. Suitable for linguistic network analysis, community detection benchmark.
A Advogato dataset of 6,551 users, 51,317 trust relationships, collected by Massa et al. (2009) via Advogato platform. Each record contains directed, weighted trust edges between users, representing explicit trust relationships expressed by users. Advogato trust network where users express weighted trust relationships, enabling research on trust propagation and social influence in online communities. Suitable for trust network analysis, influence propagation, social influence studies, trust-based recommendation systems.
Synthetic data: 150,000 generated social-media posts about Nigerian telecom operators (MTN, Airtel, Glo, 9mobile), published on Hugging Face by THUgewu as 'africa-synth'. The posts are not real user content. The file groups them into 978 rows by platform, operator, city and sentiment, with one list per field (post ID, time, text, issue category, likes, shares, comments, reach, sentiment score, operator response). Suitable for testing sentiment and customer-service pipelines, not for studying real users.
A Amazon.com movie product pages dataset of 7,911,696 movie reviews spanning 1997 to 2012, collected by McAuley & Leskovec via Amazon product review pages (2013). Each record contains product ID, reviewer user ID, reviewer profile name, helpfulness votes (numerator/denominator), star rating (1-5), Unix timestamp, review summary, and full review text. One of the largest Amazon review datasets publicly available — 8 million movie reviews with full text, helpfulness votes, and star ratings spanning 15 years of Amazon movie reviews.. Suitable for sentiment analysis, review helpfulness prediction, recommendation systems, user expertise modeling, longitudinal review behavior analysis.
A Amazon.com (product reviews and ratings) dataset of 3,376,972 nodes and 5,743,258 edges, collected by J. McAuley & J. Leskovec via Amazon product pages. Each record contains directed: False, weighted: True, temporal: True. Large bipartite user-product rating network with timestamps and 1-5 star ratings — 5.8M edges across 2.1M users and 1.2M products.. Suitable for recommender systems, rating prediction, bipartite network analysis, e-commerce research.
A Amazon dataset of 9,011 QA pairs in Appliances, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Appliances. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 21,262 QA pairs in Arts Crafts & Sewing, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Arts Crafts & Sewing. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 89,923 QA pairs in Automotive, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Automotive. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 28,933 QA pairs in Baby, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Baby. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 42,422 QA pairs in Beauty, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Beauty. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 85,865 QA pairs in Cell Phones & Accessories, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Cell Phones & Accessories. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 22,068 QA pairs in Clothing Shoes & Jewelry, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Clothing Shoes & Jewelry. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 314,263 QA pairs in Electronics, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Electronics. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 19,538 QA pairs in Grocery & Gourmet Food, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Grocery & Gourmet Food. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 80,496 QA pairs in Health & Personal Care, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Health & Personal Care. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 184,439 QA pairs in Home & Kitchen, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Home & Kitchen. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 12,136 QA pairs in Industrial & Scientific, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Industrial & Scientific. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 23,322 QA pairs in Musical Instruments, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Musical Instruments. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 43,608 QA pairs in Office Products, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Office Products. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 59,595 QA pairs in Patio Lawn & Garden, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Patio Lawn & Garden. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 36,607 QA pairs in Pet Supplies, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Pet Supplies. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 10,636 QA pairs in Software, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Software. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 146,891 QA pairs in Sports & Outdoors, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Sports & Outdoors. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 101,088 QA pairs in Tools & Home Improvement, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Tools & Home Improvement. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 51,486 QA pairs in Toys & Games, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Toys & Games. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 13,307 QA pairs in Video Games, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Amazon product Q&A data for Video Games. Suitable for question answering, sentiment analysis, product understanding, conversational AI.
A Amazon dataset of 1,396,896 QA pairs across 21 categories, collected by Wan and McAuley via Amazon API. Each record contains question, answer, question type, answer type, ASIN, timestamp. Comprehensive Amazon product Q&A data across 21 categories. Suitable for question answering, sentiment analysis, product understanding, conversational AI. The record count covers the single-answer files (qa_*.json.gz, one question and answer per line). The package also holds the multiple-answer release used in Wan and McAuley's ICDM 2016 paper (icdm/QA_*.json.gz, 17 categories: 172,617 products, 1,025,628 questions and 4,019,744 answers, with asker and answerer IDs and helpfulness votes), and McAuley's data page (index.html). Both are written as Python literals, not strict JSON (read them with ast.literal_eval).
A Anybeat dataset of 12,645 users, 49,132 connections, collected by Fire et al. (2012) via Anybeat platform. Each record contains undirected edges between Anybeat users representing social connections in the community platform. Anybeat social network capturing user connections in a public gathering place where people interact with others from around their neighborhood or across the world. Suitable for link prediction, social network analysis, community detection, online community behavior studies.
A University Rovira i Virgili (email communication network) dataset of 1,133 nodes and 5,451 edges, collected by R. Guimerà et al.. Each record contains directed: False, weighted: False, signed: False. Email communication network of a university — classic benchmark for community detection algorithms.. Suitable for community detection, email network analysis, graph algorithm benchmarking.
A Jazz musician collaboration (recordings 1912–1940) dataset of 198 nodes and 2,742 edges, collected by P. Gleiser & L. Danon. Each record contains directed: False, weighted: False, signed: False. Collaboration network of jazz musicians who recorded together — classic small-world benchmark.. Suitable for community detection, small-world networks, graph algorithm benchmarking.
A C. elegans metabolic network dataset of 453 nodes and 2,040 edges, collected by H. Jeong et al.. Each record contains directed: False, weighted: True, signed: False. Metabolic network of C. elegans — classic biological network benchmark.. Suitable for biological network analysis, community detection, scale-free network research.
A PGP (Pretty Good Privacy) key signing network dataset of 10,680 nodes and 24,316 edges, collected by M. Boguñá et al.. Each record contains directed: False, weighted: False, signed: False. Web of trust network from PGP key signings — nodes are PGP users, edges are key signatures.. Suitable for trust network analysis, community detection, graph algorithm benchmarking.
A Internet autonomous system topology (CAIDA, 2007) dataset of 26,475 nodes and 53,381 edges, collected by CAIDA via BGP routing data. Each record contains directed: False, weighted: False. Internet autonomous system (AS) peering network — nodes are AS networks, edges are peering relationships.. Suitable for internet topology analysis, network robustness, community detection.
A Internet router-level topology (CAIDA Skitter project) dataset of 1,696,415 nodes and 11,095,298 edges, collected by CAIDA Skitter project via traceroute. Each record contains directed: False, weighted: False. Large internet router topology from the CAIDA Skitter project — 1.7M nodes and 11M edges.. Suitable for internet topology analysis, network robustness, scale-free network research.
A Twitter dataset of 4,952 tweets (1,238 rumour, 3,714 non-rumour) from the 2013 Boston Marathon bombings, collected by Zubiaga, Liakata, and Procter — Augmented PHEME (Zenodo, 2019). Each record contains tweet text, binary rumour/non-rumour label, timestamp, reaction thread structure linking source tweets to replies. Augmented extension of PHEME covering the Boston Marathon bombings event with both rumour threads and non-rumour threads, providing a balanced benchmark for rumour detection with reaction graph structure. Suitable for rumour detection, misinformation classification, crisis communication analysis, stance detection, conversational thread modelling.
A Barrapunto (the Spanish Slashdot) dataset of 7,724 discussion threads, 2005-2007, from the Online Conversation Threads repository of Aragón et al. (ICWSM 2017): 172,132 nodes (articles and comments) and 169,825 comment-to-parent reply edges, with comment text, moderation scores, timestamps and thread depth. Suitable for online discourse analysis, conversation dynamics and hierarchical reply-tree structure.
A BibSonomy academic social bookmarking platform dataset of 210,467 nodes and 453,987 edges, collected by KONECT via BibSonomy API. Each record contains directed: False, weighted: True, temporal: True. Temporal bipartite network of BibSonomy users and publications they bookmarked, with Unix timestamps.. Suitable for social bookmarking analysis, academic recommendation, temporal bipartite networks.
Wikipedia
4chan
Academia.edu
Advogato
Facebook
Instagram
Twitter (X)
Amazon
The internet
Barrapunto
BibSonomy