Source-linked AI summary
Kek, Cucks, and God Emperor Trump: A Measurement Study of 4chan's Politically Incorrect Forum and Its Effects on the Web
Gabriel Emile Hine, Jeremiah Onaolapo, Emiliano De Cristofaro, Nicolas Kourtellis, Ilias Leontiadis, Riginos Samaras, Gianluca Stringhini, Jeremy Blackburn
TL;DR
The paper addresses limited scientific knowledge about 4chan’s user base, content, and effects on the wider Web by conducting a multi-faceted longitudinal study of /pol/. Across 8M posts and related platform data, it characterizes global activity, content, hate speech, and evidence of coordinated effects on YouTube.
Problem
Scientific understanding of 4chan remains limited despite /pol/’s links to the alt-right, hate speech, and broader online activity.
Method
The authors analyze /pol/ longitudinally using 8M posts from over 216K threads collected over 2.5 months, alongside linked-platform data.
Results
Evidence indicates synchronized /pol/ and YouTube activity, increased hateful comments as synchronization lag approaches zero, and likely coordination among videos sharing commenters.
Takeaways & Limitations
The study provides the first large-scale measurement of /pol/ and insight into hate and extremism trends across social media.
Takeaways & Limitations
Hate-word usage can be context-dependent, country flags can be distorted by VPNs or proxies, and the study does not classify raids.
Abstract
from arXiv · showhide
The discussion-board site 4chan has been part of the Internet's dark underbelly since its inception, and recent political events have put it increasingly in the spotlight. In particular, /pol/, the "Politically Incorrect" board, has been a central figure in the outlandish 2016 US election season, as it has often been linked to the alt-right movement and its rhetoric of hate and racism. However, 4chan remains relatively unstudied by the scientific community: little is known about its user base, the content it generates, and how it affects other parts of the Web. In this paper, we start addressing this gap by analyzing /pol/ along several axes, using a dataset of over 8M posts we collected over two and a half months. First, we perform a general characterization, showing that /pol/ users are well distributed around the world and that 4chan's unique features encourage fresh discussions. We also analyze content, finding, for instance, that YouTube links and hate speech are predominant on /pol/. Overall, our analysis not only provides the first measurement study of /pol/, but also insight into online harassment and hate speech trends in social media.
1 Introduction
The paper studies /pol/ as an influential but underexamined space where fringe political communities generate culture, hate, and activity beyond 4chan. Using a longitudinal, multi-faceted dataset, it characterizes users, content, and effects on other platforms.
- Motivation: 4chan combines anonymity and ephemerality with substantial influence over Internet culture, memes, the alt-right, and harassment.Threads are periodically pruned, while the site has also hosted hate speech, pornography, trolling, murder confessions, and attacks on other sites.
- Study scope: The study analyzes /pol/ using 8M posts from over 216K conversation threads collected over 2.5 months.It examines posting behavior, shared links and images, hate speech, country-level topics, and effects on other platforms.
- Study scope: /pol/ users are distributed around the world, and 4chan’s features encourage fresh discussions despite anonymous posting.The paper also reports many different voices among posters.
- Study scope: YouTube links, right-wing news sources, unique images, hate speech, and cross-platform “raids” are central topics of the analysis.The study measures raids as activity affecting conversations on other services, especially YouTube.
2 4chan
4chan is an anonymous imageboard organized into topic-focused boards, where ephemeral threads support rapid, loosely moderated discussion. The paper focuses on /pol/ and compares it with /sp/ and /int/.
- Imageboard model: Users create threads with an image, while replies can include images, references, and quoted text.Boards are organized around particular interests, such as politics, sports, or international topics.
- Anonymity: Anonymity means users need no account and posts carry no persistent identity, although tripcodes and poster IDs can link activity in limited ways.Poster IDs operate within a thread, while tripcodes can connect threads across time through pseudo-identity.
- Flags: Flags identify the posting country through IP geolocation, but VPNs and proxies can manipulate this signal.The feature is used on /pol/, /sp/, and /int/.
- Ephemerality: Threads are periodically pruned through a bumping system, with bump and image limits preventing indefinite survival.New posts normally move threads upward, while threads exceeding board-dependent limits stop being bumped.
- Moderation: Moderation is generally lax on /pol/, with volunteer janitors able to prune content and recommend bans.More senior employees make final banning decisions.
3 Related Work
Prior research on 4chan and related platforms provides limited evidence about anonymous behavior, aggression, and content moderation. This paper extends that work by measuring /pol/ content and its effects beyond the board.
- 4chan research: The only prior 4chan measurement study examined /b/ using 5.5M posts from almost 500K threads collected over two weeks.It focused on anonymity, ephemerality, and thread types rather than cross-platform effects.
- 4chan research: The earlier /b/ study found over 90% of posts were anonymous and reported a median thread lifetime of 3.9 minutes.Its manual labeling identified 7% of posts as calls for action, including raiding behavior.
- Related platforms: Research comparing anonymous and non-anonymous platforms has found linguistic differences and studied anonymity choices, interaction patterns, and aggression.Examples include Whisper versus Twitter classifiers and analyses of Quora and Ask.fm behavior.
- Hate speech research: Other work has developed hate-speech detectors and predicted antisocial behavior on mainstream sites, but this paper focuses on /pol/.The authors distinguish their work from studies centered on supervised detection or broad comment-section behavior.
4 Datasets
The dataset combines systematic 4chan crawling with archived thread retrieval and supplementary YouTube and Twitter data. Ethical procedures addressed anonymity and the potentially offensive nature of /pol/ content.
- 4chan collection: The crawler retrieved /pol/’s thread catalog every 5 minutes and downloaded full archived threads after they were purged.The API supplied post numbers, authors, timestamps, contents, and image metadata, but the crawler did not save images.
- Supplementary data: The main dataset runs through September 12, 2016, while the raids analysis uses data through September 25 and a Twitter comparison uses 60,040,275 tweets.The Twitter sample covers September 18 to October 5, 2016.
- Ethics: The study received ethics approval, encrypted data at rest, and made no attempt to de-anonymize users.The authors note that links from 4chan to other services could potentially enable de-anonymization.
- Ethics: The researchers did not censor offensive content, warning that the paper contains language likely to be upsetting.This choice was intended to support comprehensive analysis of /pol/.
5 General Characterization
/pol/ participation is internationally diverse and supports frequent, fresh discussions shaped by board-specific bump limits. Despite anonymity, pseudo-identifiers and reply patterns reveal varied engagement, while the reply-only interaction system may reward inflammatory posting.
- Posting activity: New Zealand, Canada, Ireland, Finland, and Australia lead /pol/ in threads per Internet-using capita, while many non-English-speaking countries are represented.The United States leads in total thread creation, but participation is internationally diverse.
- Posting activity: 38.4, 57.1, and 82.9 are the mean posts per thread on /pol/, /int/, and /sp/, respectively, with medians of 7.0, 12.0, and 12.0.All three distributions are right-skewed, indicating a small number of substantially longer threads.
- Posting activity: Bump limits keep fresh content available by causing extremely popular threads to be purged sooner than the overall thread-length distribution would imply.The limits are board-dependent and eventually prevent further posts from bumping a thread.
- Posting activity: Deleted /pol/ threads have a median of around 20 posts, compared with 7 for successfully archived threads.The study identifies deleted threads through catalog observations followed by 404 responses from the archive.
- Tripcodes, poster IDs, and replies: 188,849 posts contain tripcodes, whose usage shows that 25% of tripcodes are used once and that tripcodes do not necessarily represent unique users.Tripcodes provide pseudo-identity across time, but users can employ multiple tripcodes.
- Tripcodes, poster IDs, and replies: 57% of /pol/ posts receive no direct reply, yet /pol/ averages 0.83 replies per post and has greater reply variability than /int/ or /sp/.Because replies are 4chan's only direct interaction and validation mechanism, users may craft inflammatory or controversial posts to attract them.
6 Analyzing Content
The paper analyzes /pol/ content across links, images, hate speech, and geographic patterns. It finds extensive YouTube linking, substantial production of unique images, and measurable but geographically uneven hate-speech prevalence.
- Links: YouTube is /pol/'s most-linked website, with over an order of magnitude more URLs than Wikipedia and Twitter.Archive.is, Wikileaks, pastebin, and DonaldJTrump.com also rank among the most popular domains, while major news sites appear outside the top ten.
- Images: 1,003,785 unique images comprised 45% of 2,210,972 posted images during the observation period.About 70% of images appeared once, nearly 95% no more than five times, and the most popular image appeared 838 times.
- Images: Over 1M unique images were posted in 2.5 months, most either original or sourced from outside /pol/.The authors connect this continual production of new content with /pol/'s position at the heart of the Internet hate movement.
- Hate speech: 12% of /pol/ posts contained hateful terms, compared with 6.3% on /sp/, 7.3% on /int/, and 2.2% in the sampled tweets.The authors identify hateful posts using a manually refined Hatebase dictionary and NLTK word-form matching.
- Hate speech: “Nigger” appeared in more than 2% of posts, while “faggot” and “retard” each appeared in over 1%.The first term appeared in 265K posts, approximately 120 posts per hour; usage dropped sharply after the top three words.
- Country analysis: Country-level hate-speech prevalence ranged from around 4.15% to around 30%, while most of 239 countries fell between 8% and 12%.The country heat map covers countries with at least 1,000 posts; proxy use may overrepresent some high-prevalence countries.
7 Raids Against Other Services
The paper examines how /pol/ raids target YouTube and Twitter, combining case-study evidence with activity synchronization analysis. Operation Google spread rapidly within /pol/ but had limited Twitter prevalence, while synchronized /pol/ and YouTube activity was associated with hate speech and commenter overlap.
- Raids Against Other Services: Raids are semi-organized attempts to disrupt another site's community through coordinated content rather than network-level attacks.Calls to action typically link a target and prompt users to harass it; threads can aggregate reactions and sock-puppet accounts.
- Case Study: “Operation Google”: Operation Google replaced racially charged words with company names to evade anti-trolling systems, producing a rapid burst of activity within /pol/.On September 22, “google” appeared at over five times its normal rate and “Skype” at almost double; usage had subsided by September 26.
- Case Study: “Operation Google”: Operation Google hashtags appeared on Twitter beginning September 22, but #dumbgoogles appeared in only 5 of 3M tweets (0.00016%) on the most active day.The observed hashtag activity peaked on September 23, yet the impact was much more prevalent on /pol/ than on Twitter.
- Activity Modeling: The study examines 19,568 YouTube videos linked by 10,809 /pol/ threads, looking for elevated and synchronized commenting activity rather than explicit raid calls.The analysis expects hateful-comment rates to increase after a YouTube link is posted on /pol/.
- Activity Modeling: Fourteen percent of linked YouTube videos experienced a commenting peak during the period they were discussed on /pol/, although this does not by itself establish that a raid occurred.The analysis uses normalized thread lifetimes and considers YouTube comments within a normalized [-10, +10] period, containing 35% of comments in the dataset.
- Evidence of Raids: YouTube comments containing more hate speech were significantly more synchronized with /pol/ activity, with the comparison yielding p < 0.01 in a 2-sample Kolmogorov-Smirnov test.The reported relationship links higher hateful-comment rates with shorter synchronization lags, and synchronization also correlates with commenter overlap.
8 Discussion & Conclusion
The study characterizes /pol/ across activity, content, and cross-platform effects, finding distinctive posting behavior, globally distributed participation, abundant original content, and evidence consistent with raids on YouTube. It also identifies limits around contextual hate-speech interpretation, geographic inference, and raid classification.
- General characterization: Distinct posting behaviors across /pol/, /sp/, and /int/ show that bump limits consistently promote fresh content.Country-flag data indicate that Americans dominate in absolute numbers, while many other countries are well represented in posts per capita.
- Content: 70% of 1M unique images were posted once and 95% fewer than five times, indicating that /pol/ contains highly original content.The authors suggest this ability to find or produce original content may contribute to /pol/'s perceived centrality in online hate.
- Content: The majority of /pol/ links point to YouTube, while tabloid and right-wing news links substantially outnumber links to mainstream sites.These findings describe the board's platform and news-source preferences.
- Cross-platform effects: YouTube commenting peaks tend to occur during linked /pol/ thread lifetimes, and synchronization increases as the lag approaches zero.The authors interpret this temporal alignment as quantitative evidence that raids take place, without claiming they can classify individual raids.
- Conclusion: The study offers a foundation for future research on fringe groups, hate speech, and online harassment campaigns.The authors frame the work as a first measurement study and as insight into growing hate and extremism trends on social media.
- Limitations: Hate-word analysis is limited because word usage depends on context, including sarcasm and trolling, which the study leaves for future work.The country analysis may also be influenced by VPNs or proxies, motivating deeper analysis of language and posting behavior.
A Rare Pepes
This section presents a collection of rare Pepe images, ranging from modern and national-themed depictions to political, artistic, commercial, and Trump-related reinterpretations.
- A Rare Pepes: The collection includes a modern Pepe depicting the historic rise of /pol/ and a beret-wearing Pepe with a cigarette and accordion.
- A Rare Pepes: It includes a CNN-commissioned Pepe commemorating Pepe's recognition as a hate symbol and an ultra-rare Pepe eating a Publix Deli Sub Sandwich.
- A Rare Pepes: Other images depict Hillary Clinton, Julian Assange carrying Democratic National Convention secrets, and a reinterpretation of Goya's “Saturn Devouring His Son.”
- A Rare Pepes: The section also shows a very comfy Pepe, a mischievous witch Pepe, and the now-iconic Trump Pepe.