[Paper Review] Racism is a Virus: Anti-Asian Hate and Counterhate in Social Media during the COVID-19 Crisis

Caleb Ziems, Bing He+2 more|arXiv (Cornell University)|May 25, 2020

Hate Speech and Cyberbullying DetectionComputer Science78 references83 citations

TL;DR

This study analyzes anti-Asian hate and counterhate speech on Twitter during the COVID-19 pandemic using COVID-HATE, a dataset of 30+ million tweets and 87+ million nodes. It identifies that hate is contagious, bots amplify hate, and counterhate speech effectively reduces the likelihood of users turning hateful, offering a data-driven approach to mitigating online racism during crises.

ABSTRACT

The spread of COVID-19 has sparked racism, hate, and xenophobia in social media targeted at Chinese and broader Asian communities. However, little is known about how racial hate spreads during a pandemic and the role of counterhate speech in mitigating the spread. Here we study the evolution and spread of anti-Asian hate speech through the lens of Twitter. We create COVID-HATE, the largest dataset of anti-Asian hate and counterhate spanning three months, containing over 30 million tweets, and a social network with over 87 million nodes. By creating a novel hand-labeled dataset of 2,400 tweets, we train a text classifier to identify hate and counterhate tweets that achieves an average AUROC of 0.852. We identify 891,204 hate and 200,198 counterhate tweets in COVID-HATE. Using this data to conduct longitudinal analysis, we find that while hateful users are less engaged in the COVID-19 discussions prior to their first anti-Asian tweet, they become more vocal and engaged afterwards compared to counterhate users. We find that bots comprise 10.4% of hateful users and are more vocal and hateful compared to non-bot users. Comparing bot accounts, we show that hateful bots are more successful in attracting followers compared to counterhate bots. Analysis of the social network reveals that hateful and counterhate users interact and engage extensively with one another, instead of living in isolated polarized communities. Furthermore, we find that hate is contagious and nodes are highly likely to become hateful after being exposed to hateful content. Importantly, our analysis reveals that counterhate messages can discourage users from turning hateful in the first place. Overall, this work presents a comprehensive overview of anti-Asian hate and counterhate content during a pandemic. The COVID-HATE dataset is available at this http URL.

Motivation & Objective

• To understand the dynamics of anti-Asian hate speech and counterhate on social media during the COVID-19 pandemic.
• To investigate how hateful and counterhateful users differ in engagement, virality, and network behavior.
• To examine the role of bots in amplifying hate and counterhate content.
• To assess whether exposure to counterhate speech reduces the likelihood of users adopting hateful behavior.
• To create and release a large-scale, hand-labeled dataset (COVID-HATE) for future research on online hate during public health crises.

Proposed method

• Constructed the COVID-HATE dataset, comprising over 30 million tweets and 87 million nodes from Twitter, spanning three months during the pandemic.
• Created a hand-labeled dataset of 2,400 tweets to train a text classifier for hate and counterhate detection, achieving an average AUROC of 0.852.
• Applied longitudinal analysis to compare user engagement and behavioral shifts before and after users posted their first hate or counterhate tweet.
• Identified and classified bot accounts among hateful and counterhateful users, comparing their virality and follower growth.
• Mapped the social network of hate and counterhate users to analyze interaction patterns and content contagion.
• Trained and evaluated a text classification model using supervised learning on hand-labeled data to detect hate and counterhate content at scale.

Experimental results

Research questions

• RQ1 How do hateful and counterhateful users differ in their engagement and virality patterns before and after posting hate content?
• RQ2 To what extent do bots contribute to the amplification of anti-Asian hate and counterhate on Twitter?
• RQ3 Are hateful and counterhateful users isolated in polarized communities, or do they interact across ideological lines?
• RQ4 Is exposure to counterhate content associated with a reduced likelihood of users becoming hateful?
• RQ5 How contagious is hate in online social networks, and what role does network structure play in its spread?

Key findings

• Hateful users became significantly more engaged and vocal on Twitter after posting their first anti-Asian hate tweet, compared to counterhate users.
• Bots made up 10.4% of hateful users and were more vocal and hateful than non-bot users, with hateful bots attracting more followers than counterhate bots.
• Hateful and counterhate users frequently interacted with each other, indicating they do not exist in isolated, polarized communities.
• Exposure to hateful content increased the likelihood of users adopting hateful behavior, demonstrating that hate is contagious in online networks.
• Counterhate messages were effective in reducing the likelihood of users turning hateful, suggesting a protective effect against radicalization.
• The study identified 891,204 hate tweets and 200,198 counterhate tweets within the COVID-HATE dataset, providing a rich resource for future research.