Skip to content
GUIDEUnderstanding fake news, misinformation & disinformation

2019 UK General Election Twitter Dataset

Language: English
Get the data

Overview

A large-scale Twitter dataset collected during the 2019 UK General Election period, comprising 11.2 million tweets from 1.2 million distinct users. The dataset is specifically designed for studying coordinated online behavior, political polarization, and computational propaganda during electoral campaigns. It includes user interactions, hashtag usage, political leaning labels inferred from hashtag polarity, and community membership labels from coordinated behavior analysis.

Data

Size: 11.2M tweets, 1.2M users

Collection period: November 12 – December 12, 2019 (1 month pre-election window)

Language: English

Platform: Twitter (via Twitter Streaming API)

Collection method: Hashtag-based filtering using 13 curated political hashtags:

  • GE2019, #GeneralElection19

  • VoteLabour, #saveJeremy, #labourmanifesto, #ClinimateChange

  • BackBoris, #VoteConservative, #ToriesInForTheLongGame

  • VoteTactically, #RevokeArticle50, #stopbrexit, #ForTheMary

Data types included:

  • Original tweets
  • Retweets
  • Replies
  • User metadata
  • Interaction metadata (likes, retweets, replies)

Labels

Political leaning labels

Computed via label propagation algorithm applied to hashtag polarity:

  • Labour (red) — left-leaning hashtags
  • Conservative (blue) — right-leaning hashtags
  • Neutral/Tactical voting (orange) — anti-establishment, pro-liberal-democrat hashtags
  • Scottish National Party (yellow) — Scottish independence-focused hashtags

Coordinated community labels

Identified through iterative community detection at increasing coordination thresholds:

  1. CON — Conservative users (coordination extent ~0.75–0.9)
  2. LAB — Labour users (coordination extent ~0.7–0.85)
  3. TVT — Neutral/tactical voting users (coordination extent ~0.3–0.5)
  4. SNP — Scottish National Party supporters (coordination extent ~0.4–0.6)
  5. B60 — Back60 activist initiative users (small, tight-knit cluster)
  6. ASE — Anti-establishment Conservative supporters
  7. LCH — Anti-semitism activist cluster

Automation scores

Botometer scores provided for user-level automation classification (distinguishing human-operated vs. bot accounts).

Format

Data files:

  • Tweets table (CSV/JSON): tweet ID, user ID, created_at, text, in_reply_to_user_id, in_reply_to_status_id, retweet_count, favorite_count, hashtags, user_mentions, URLs
  • User table (CSV/JSON): user ID, name, screen_name, followers_count, friends_count, statuses_count, political_leaning_label, community_label, botometer_score
  • Network files: user similarity networks (edge lists with weights)
  • Hashtag metadata: hashtag frequency, polarity scores, community association

Format options: CSV, JSON, edge lists for network analysis

Availability

Repository: https://github.com/iit-cnr/GUIDE-2019UK-GE-Twitter-Dataset

The dataset is made publicly available subject to Twitter's terms of service and research-use policies. Full raw tweets are provided via tweet ID lists; user should hydrate using Twitter API or archive.org. Some user/tweet deletions may occur over time (users delete accounts, tweets removed).

Applications

This dataset supports research on:

  • Coordinated inauthentic behavior detection: network-based methods for detecting coordinated political campaigns
  • Political polarization on social media: characterizing structural polarization during electoral events
  • Computational propaganda strategies: analyzing different political factions' use of automation, coordination, and messaging
  • Election-period information dynamics: temporal analysis of how political narratives spread and evolve
  • Community structure in political networks: methodological development for community detection with coordination awareness
  • Cross-platform coordination: groundwork for studying how coordination extends beyond Twitter to other platforms
  • FakeNewsNet — social media news dataset with engagement and stance features
  • Twitter 2016 U.S. Election Dataset — related election-period Twitter collection
  • LIAR — fact-checked claim dataset (complementary for veracity labeling)

Notes

Strengths:

  • Large scale (11.2M tweets) enables robust statistical analysis
  • Clear temporal boundaries (1-month election window) facilitate event-level studies
  • Political leaning labels enable polarization analysis
  • Community labels from coordinated behavior analysis allow downstream research on coordination patterns
  • Public availability supports reproducibility and comparative studies

Limitations:

  • Twitter-only; does not capture coordinated behavior on other platforms
  • Collection period ends before election day; does not capture post-election activity or outcome effects
  • Political leaning labels derived algorithmically from hashtags; may misclassify ambiguous or multi-topic users
  • User deletions and tweet removals over time will affect availability
  • No ground-truth validation of community labels (automated detection may have false positives/negatives)

Ethical considerations:

  • Contains user-generated content with potential for re-identification; researchers must follow Twitter ToS
  • Political classification of users should not be used to make individual-level inferences or targeted interventions
  • Dataset includes potential offensive language and political attacks; researchers must handle sensitively

Papers using this dataset

Detection methodsBots & coordinated campaigns Coordinated Behavior on Social Media in 2019 UK General Election Nizzoli et al. · arXiv · 2020