Abstract
With the world-wide development of 2019 novel coronavirus, although WHO has officially announced the disease as COVID-19, one controversial term - "Chinese Virus" is still being used by a great number of people. In the meantime, global online media coverage about COVID-19-related racial attacks increases steadily, most of which are anti-Chinese or anti-Asian. As this pandemic becomes increasingly severe, more people start to talk about it on social media platforms such as Twitter. When they refer to COVID-19, there are mainly two ways: using controversial terms like "Chinese Virus" or "Wuhan Virus", or using non-controversial terms like "Coronavirus". In this study, we attempt to characterize the Twitter users who use controversial terms and those who use non-controversial terms. We use the Tweepy API to retrieve 17 million related tweets and the information of their authors. We find significant differences between these two groups of Twitter users across their demographics, user-level features like the number of followers, political following status, as well as their geo-locations. Moreover, we apply classification models to predict Twitter users who are more likely to use controversial terms. To our best knowledge, this is the first large-scale social media-based study to characterize users with respect to their usage of controversial terms during a major crisis.
Introduction
As COVID-19 spread globally in early 2020, social media became a primary venue for "platformed racism"—xenophobic discourse amplified by digital networks. Despite official naming conventions by the World Health Organization (WHO) intended to avoid stigmatization, terms like "Chinese Virus" and "Wuhan Virus" trended heavily. While existing research often focuses on the content of such discourse, this study investigates the profiles of the users behind it. By identifying the characteristics of users who adopt controversial terminology, we aim to better understand the roots of online discrimination and inform interventions against hate speech.
Methodology
The research utilized a multi-stage pipeline to profile users based on 17 million tweets collected via the Tweepy API in March 2020. Data was categorized into a Controversial Dataset (CD) and a Non-controversial Dataset (ND) based on keyword usage.
To overcome Twitter's lack of demographic data, the Face++ API was employed to infer age and gender from user profile pictures. Political leanings were estimated based on whether users followed major 2020 U.S. presidential candidates. Geographic analysis mapped profile locations to zip codes, classifying them as urban, suburban, or rural based on population density. Finally, machine learning models, including Random Forest (RF) and XGBoost, were trained to predict user group membership based on these features.
Characterizing User Profiles
Our analysis revealed distinct profiles for users in the CD and ND groups across several dimensions.
Social Capital and Experience
Users in the non-controversial group (ND) generally possessed higher "social capital," characterized by more followers, more friends, and older accounts. This suggests that more experienced users may be more cautious with their language to protect their online reputation.
Demographic and Geographic Trends
Demographic analysis showed that the "controversial" user is statistically more likely to be male and over the age of 45. In contrast, younger users (under 35) and female users showed a strong preference for neutral, non-stigmatizing language. For instance, 21% of users in the non-controversial group were in the 18-24 age bracket, compared to only 16.5% in the controversial group.
Geographically, urban users were significantly more likely to use neutral terms, while rural and suburban users leaned toward controversial ones. This aligns with broader geographic distributions of political ideologies in the United States.
Political Polarization
Political following proved to be the most significant predictor. Following Donald Trump was strongly correlated with using controversial terms, with 21.6% of the CD group following only Trump, compared to just 2.4% of the ND group. Conversely, following Democratic candidates was associated with neutral language.
Predictive Performance
The study demonstrated that user behavior regarding controversial language is highly predictable. The Random Forest (RF) classifier emerged as the most effective model, achieving an AUC-ROC of 0.874 on the geo-location dataset and 0.833 on the demographic dataset. These results underscore the strong link between user characteristics and their choice of terminology.
Conclusion and Future Work
This research highlights how demographic, geographic, and political factors shape the adoption of controversial language during a global crisis. While the study assumes term usage implies agreement—a limitation that doesn't account for users criticizing the terms—the high predictive accuracy of the models suggests a robust correlation. Future work should incorporate Natural Language Processing (NLP) to analyze sentiment and context, distinguishing between those promoting stigmatizing language and those discussing it critically.
Summary updated