Source-linked AI summary
"Go eat a bat, Chang!": On the Emergence of Sinophobic Behavior on Web Communities in the Face of COVID-19
Fatemeh Tahmasbi, Leonard Schild, Chen Ling, Jeremy Blackburn, Gianluca Stringhini, Yang Zhang, Savvas Zannettou
TL;DR
The paper addresses limited evidence on how Sinophobic behavior emerges and evolves online during COVID-19. It analyzes large-scale Twitter and 4chan /pol/ datasets with temporal and embedding-based methods, finding increased Sinophobic content across both platforms, platform-specific shifts in language, and newly created slurs. The study’s scope is constrained by its focus on two selected Web communities and acknowledged differences in their geographic and social dynamics.
Problem
The paper addresses the limited understanding of how Sinophobic language and its context evolve online during the COVID-19 pandemic.
Method
The authors analyze large-scale Twitter and 4chan /pol/ datasets collected over five months using temporal analysis, word embeddings, and graph analysis.
Results
COVID-19 coincided with a rise in Sinophobic content on both fringe /pol/ and mainstream Twitter, while word embeddings revealed evolving and newly created Sinophobic slurs.
Takeaways & Limitations
Sinophobic dissemination is cross-platform and evolves substantially, supporting multi-platform study and tools that can track changing online behavior.
Takeaways & Limitations
The study focuses on Twitter and /pol/, whose differences may reflect geographic distribution, social distance, or /pol/’s greater susceptibility to conspiracies and racism-related posts.
Abstract
from arXiv · showhide
The outbreak of the COVID-19 pandemic has changed our lives in unprecedented ways. In the face of the projected catastrophic consequences, many countries have enacted social distancing measures in an attempt to limit the spread of the virus. Under these conditions, the Web has become an indispensable medium for information acquisition, communication, and entertainment. At the same time, unfortunately, the Web is being exploited for the dissemination of potentially harmful and disturbing content, such as the spread of conspiracy theories and hateful speech towards specific ethnic groups, in particular towards Chinese people since COVID-19 is believed to have originated from China. In this paper, we make a first attempt to study the emergence of Sinophobic behavior on the Web during the outbreak of the COVID-19 pandemic. We collect two large-scale datasets from Twitter and 4chan's Politically Incorrect board (/pol/) over a time period of approximately five months and analyze them to investigate whether there is a rise or important differences with regard to the dissemination of Sinophobic content. We find that COVID-19 indeed drives the rise of Sinophobia on the Web and that the dissemination of Sinophobic content is a cross-platform phenomenon: it exists on fringe Web communities like \dspol, and to a lesser extent on mainstream ones like Twitter. Also, using word embeddings over time, we characterize the evolution and emergence of new Sinophobic slurs on both Twitter and /pol/. Finally, we find interesting differences in the context in which words related to Chinese people are used on the Web before and after the COVID-19 outbreak: on Twitter we observe a shift towards blaming China for the situation, while on /pol/ we find a shift towards using more (and new) Sinophobic slurs.
1 Introduction
The paper examines the emergence and evolution of online Sinophobia during COVID-19 across mainstream and fringe Web communities. It argues that the Web’s role in sustaining daily life also enables misinformation, conspiracy theories, and racist rhetoric.
- 1 Introduction: The Web became essential for daily life during social distancing while also enabling misinformation, conspiracy theories, and racist rhetoric.The paper frames these harmful uses as especially concerning amid online racism and the politically charged emergence of SARS-CoV-2.
- 1 Introduction: The study analyzes how Sinophobic behavior emerged and evolved during the COVID-19 crisis on Twitter and 4chan’s /pol/.It uses temporal analysis, word embeddings, and graph analysis to compare discussions before and after the crisis.
- 1 Introduction: The paper reports increased discussions about China and Chinese people, alongside increased use of specific Sinophobic slurs, on both Twitter and /pol/ after the outbreak.These increases coincide with real-world events related to the pandemic.
- 1 Introduction: Sinophobic content appears across platforms, with important differences in slur usage between mainstream Twitter and fringe /pol/.“Chinazi” is most popular on Twitter, whereas “chink” is most popular on /pol/.
- 1 Introduction: The study warns that its Web-community material is likely to be offensive or racist because the authors do not censor language.Readers are warned that the presented content may be offensive and upsetting.
2 Related Works
Related work covers COVID-19 discussions, misinformation, online racism, racist-language measurement, and demographic predictors of racist activity. The paper positions itself as the first data-driven study of evolving anti-Chinese and anti-Asian racist rhetoric during COVID-19.
- 2 Related Works: Prior COVID-19 research examined pandemic datasets, misinformation narratives, and reactions to related policies on social media.Examples include a 50M-tweet dataset and studies of misinformation on Twitter.
- 2 Related Works: Prior racism research documented pervasive online racism, links between racist activity and hate crimes, anonymous online aggression, and users’ experiences of racism.These studies addressed prevalence, behavioral influence, geographic correlation, and self-narration.
- 2 Related Works: Related quantitative work measured racist-language evolution with word embeddings and studied demographic traits as predictors of racist activity on Twitter.The cited work found that male and younger users were more likely to engage in racism on Twitter.
- 2 Related Works: The authors describe their work as the first data-driven study of evolving racist rhetoric against Chinese people and people of Asian descent during COVID-19.The novelty claim focuses on rhetoric evolution in light of the pandemic.
3 Datasets
The study compares two large-scale Web communities: mainstream Twitter and fringe 4chan /pol/. It collects English tweets and all /pol/ posts over the same approximately five-month period to examine Sinophobic behavior.
- 3 Datasets: The researchers collect and analyze two large-scale datasets from Twitter and 4chan’s Politically Incorrect board (/pol/).The communities are selected as representative examples of mainstream and fringe Web communities.
- 3 Datasets: 222,212,841 English tweets were collected from Twitter between 28 October 2019 and 22 March 2020 using a 1% random Streaming API sample.Twitter is described as a popular mainstream microblog used to disseminate information.
- 3 Datasets: 16,808,191 posts were collected from 4chan’s /pol/ between 28 October 2019 and 22 March 2020.4chan is an anonymous imageboard organized into boards with distinct topics and moderation policies.
- 3 Datasets: Twitter represents a mainstream community, whereas /pol/ represents a fringe community known for disseminating hateful or weaponized information.This contrast motivates the cross-platform comparison.
4 Temporal Analysis
Temporal analysis shows that discussion of China and Sinophobic slurs rose around major COVID-19 events on both /pol/ and Twitter, with platform-specific timing and intensity. /pol/ showed stronger alignment between China-related discussion and slur use, while Twitter displayed different slur popularity and a stronger later peak.
- China-related terms: Discussion of “china” and “chinese” on /pol/ rose around the Wuhan lockdown, increased again as COVID-19 spread through Europe, and peaked around Italy’s lockdown and the “Chinese Virus” tweet.The first increase occurred around January 23, the rate rose again around February 23 and March 9, and a second peak appeared around March 16.
- China-related terms: Twitter showed the same overall temporal pattern, but relative discussion was lower during the initial Wuhan-related peak than after COVID-19 spread into Europe.The paper suggests geographic distribution as one possible explanation, while noting that /pol/ may be more readily inflamed by conspiracy and racism-related posts.
- Racial slurs: Sinophobic slurs on both platforms exhibited peaks around January 23 and March 16, with more than 1% of /pol/ posts containing “chink” at the highest points.The authors interpret this temporal correspondence as evidence that the COVID-19 crisis drove a rise in online Sinophobia.
- Racial slurs: Twitter’s Sinophobic slurs were more popular around mid-December 2019 than on /pol/, coinciding with a new China–United States trade deal.This timing differs from the main /pol/ peaks associated with the pandemic’s later development.
- Racial slurs: The most popular slur was “chink” on /pol/ but “chinazi” on Twitter, where “chinazi” barely appeared on /pol/.These differences indicate that the two communities used distinct Sinophobic vocabularies, limiting simple dictionary-based comparisons.
- Discussion: The paper relates intensified China-focused discussion after Western outbreaks to scapegoating, as China’s association with COVID-19 made it a target amid the spreading pandemic.The discussion accelerated once the Western world became affected, after China had already been associated with the outbreak and early containment measures.
5 Content Analysis
Word2vec and graph analyses show that COVID-19-related and Sinophobic language appears on both /pol/ and Twitter, with platform-specific contexts and slurs.
- Method: The analysis trains word2vec models on complete, weekly, and historical data to compare word contexts over time.Words used in similar ways are mapped to nearby vectors in a high-dimensional space.
- 4chan’s /pol/: On /pol/, derogatory terms such as “chink,” “chinkland,” and “chiniggers” occur among words similar to “china” and “chinese.”The four most similar words to “virus” are “coronovirus,” “covid,” “coronavirus,” and “corona.”
- Twitter: Twitter also contains Sinophobic language, but words similar to “china” and “chinese” include political terms such as “government,” “ccp,” and “chinazi.”This indicates that Sinophobic content is present in both fringe and mainstream Web communities.
- 4chan’s /pol/: The /pol/ graph contains COVID-19, bioweapon, and tightly connected communities of derogatory terms targeting Asian and Chinese people.Examples include “bioattack,” “ricenigger,” “chinksect,” “chankoro,” “chinks,” “yellowniggers,” and “pindick.”
- Twitter: The Twitter graph combines anti-China terms such as “makechinapay” and “blamechina” with virus-related terms and some support for Chinese people.Examples include “chinawuhanvirus,” “chinaflu,” “coronacontrol,” and “staystrongchina.”
- Sinophobic language: The observed profanities include racialized insults aimed at Asian people and culturally oriented slurs attacking dietary habits and skin tone.The paper relates these targeted slurs to defensive aggression and societal oppression.
6 Content Evolution
Temporal word-embedding analyses show that Sinophobic language and COVID-19-related associations intensified after the outbreak, with distinct shifts on /pol/ and Twitter.
- Content Evolution: The paper addresses how Sinophobic language changes over time, including shifts in word context and the creation of new extremist slang.These dynamics matter because online extremist language can use community-specific memes and terms absent from ordinary usage.
- Evolution on 4chan’s /pol/: Before the outbreak, /pol/ already associated “china” and “chinese” with racial slurs, while “virus” was linked mainly to general diseases and outbreaks.“Chinks” and “chink” appear among similar terms in the early models.
- Evolution on 4chan’s /pol/: Later /pol/ models show stronger associations between China-related terms and slurs, bioterrorism, bioweapon narratives, and sarcastic racialized language.The final graph contains tightly connected hateful communities with terms such as “ricenigger,” “zipperhead,” “bugpeople,” and “subhumans.”
- Emergent language: New terms emerged on both platforms around late January 2020, including “batsoup,” “biolab,” “biowarfare,” “chinavirus,” and “wuhanpneumonia.”The paper links these emergent terms to discussions of the outbreak’s alleged origins and conspiracy theories.
- Evolution on Twitter: Twitter shifts from political and disease-related contexts toward COVID-19-linked Sinophobic terms such as “chinazi,” “chinavirus,” and “kungflu.”Later political terms include “ccpvirus” and “boycottchina,” conveying revenge or punishment toward China.
- Semantic Changes between Words: Cosine similarity between “chinese” and “virus” rose above 0.5 after the week ending January 19, 2020, reaching over 0.6 in the last model.Similarity between “chinese” and “chink” also increased on both platforms, with a larger increase on Twitter.
7 Conclusion
The conclusion argues that COVID-19 made the Web both essential for social connection and a venue for increasing Sinophobic language across mainstream and fringe communities. It calls for multi-platform monitoring and tools to understand evolving hateful behavior and mitigate potential real-world violence.
- The Web became more essential for information, communication, and socialization during quarantines, but also spread hate speech targeting Chinese people.
- COVID-19 increased Sinophobic content on both /pol/ and Twitter, while many new Sinophobic slurs emerged as the crisis progressed.
- Sinophobic content is a cross-platform phenomenon spanning fringe communities such as /pol/ and mainstream platforms such as Twitter.
- The authors recommend a multi-platform perspective and new techniques to track behavioral change and develop counter-measures against potential real-world violence.
- The study presents the COVID-19 crisis as an opportunity to examine hateful-language evolution while calling for action against its proliferation.