Social bots

Social bots are automated software agents that interact with users on social media platforms, generating content, engaging in discussions, following accounts, and sharing content. Bots range from benign utility services (news feeds, weather alerts, reminder services) to sophisticated actors designed to manipulate discourse, spread misinformation, and influence elections.

Bot characteristics

Benign bots: - News aggregators: Automatically post headlines from RSS feeds - Reminder services: Schedule-based notifications - Customer service: Automated support interactions - Event notifications: Conference alerts, package tracking

Malicious bots: - Amplification bots: Automatically retweet or share content to inflate visibility - Propaganda bots: Systematically spread false or misleading narratives - Troll bots: Harassment campaigns targeting individuals or groups - Spam/scam bots: Phishing, commercial spam, cryptocurrency scams - Infiltration bots: Acquire followers and credibility to establish influence

Behavioral signatures

Bots typically exhibit distinctive behavioral patterns: - High retweet rates relative to tweet generation - Periodic posting patterns (Poisson process-like timing) - Longer usernames on average - More mentions and hashtags in tweets - Distinct sentiment profiles (often more negative) - Rapid account age growth when first active - Concentrated activity during specific hours

Sophisticated bots increasingly mimic human temporal signatures and content generation patterns, blurring the boundary between bot and human behavior.

Role in information ecosystems

Amplification: Bots dramatically increase reach and visibility of coordinated messages. During crises (elections, pandemics, conflicts), bot networks can trend topics artificially and shape public perception.

Coordinated campaigns: Organized actors (state sponsors, political campaigns, commercial entities) deploy coordinated bot networks to: - Trend false narratives - Attack political opponents - Harass journalists and activists - Manipulate financial markets - Game search rankings and recommendation algorithms

Polarization: Bot networks can amplify divisive content, attack opposing viewpoints, and reinforce echo chambers.

Detection and challenges

See Bot detection for comprehensive coverage of detection methods (network analysis, feature-based classifiers, crowd-sourcing) and associated challenges (concept drift, sophisticated bot design, false positive tradeoffs).

Key papers in this wiki

• Botometer 101: Social bot practicum for computational social scientists — Tutorial on Botometer, the most widely-used machine learning tool for detecting social bots on Twitter; explains feature engineering (1,000+ features), model variants (V4 and Lite), practical guidance on score interpretation and responsible use
• A Decade of Social Bot Detection — Decade-long review of bot detection research (2010–2020); longitudinal analysis showing shift from individual-account to group-level detection; documents bot evolution across three waves; catalogs 230+ detection techniques
• Detection of Novel Social Bots by Ensembles of Specialized Classifiers — Ensemble of specialized classifiers for detecting novel social bots; shows heterogeneous bot behaviors require different feature sets; achieves 56% improvement in cross-domain F1 score; deployed in Botometer v4
• Scalable and Generalizable Social Bot Detection through Data Selection — Scalable metadata-only bot detection using only 20 user features; systematic analysis of 13 labeled datasets reveals contradictions and separability patterns; shows data selection (training on curated subset) improves cross-domain generalization better than exhaustive training
• Arming the public with artificial intelligence to counter social bots — Reviews bot types, impact, and detection methods; empirical study of how users interact with Botometer bot detection tool; proposes calibration methods to make bot scores interpretable
• Fagni et al. (2020) — TweepFake: about detecting deepfake tweets — Text-based detection of machine-generated bot tweets; TweepFake dataset of 25,572 human vs. bot-generated tweets; benchmarks 13 detection methods; fine-tuned transformers achieve 90% accuracy on detecting older generation methods, but only 65–80% on GPT-2 generated content
• Varol et al. (2017) — Online Human-Bot Interactions: Detection, Estimation, and Characterization — Large-scale framework for bot detection; develops 1,150-feature supervised classifier; evaluates on 14M accounts; estimates bot prevalence (9–15%); identifies behavioral phenotypes through clustering
• Ferrara et al. (2015) — The Rise of Social Bots — Foundational survey of social bots; characterizes bot phenomenon; proposes taxonomy of detection approaches (network-based, crowd-sourced, feature-based); discusses arms race between bot sophistication and detection methods
• Stella, Ferrara & De Domenico (2018) — Bots increase exposure to negative and inflammatory content in online social systems — large-scale analysis showing bots strategically amplify inflammatory content and target human influencers

Connections

• Bot detection — technical approaches and challenges
• Information operations — state-sponsored bot campaigns
• Coordinated inauthentic behavior — broader category
• Twitter dynamics — primary platform for bot research
• Social media and misinformation — bots as amplification vector