Abstract
Social media censorship in China is a complex and dynamic process. While much research has focused on the content that is removed, this study investigates the characteristics of the users who share such content. By analyzing over 5,000 unique Weibo users and their repost histories, we identify key demographic and behavioral patterns. Our findings indicate that users with higher social capital and verified accounts exhibit greater caution, while users in economically developed regions and specific political contexts show a higher propensity for sharing sensitive information.
Introduction
Weibo, often described as China's equivalent to Twitter, is a primary hub for public discourse but is subject to rigorous censorship. Posts related to sensitive political topics, social unrest, or unapproved news are frequently deleted by platform moderators. Understanding the profile of users who engage with this "risky" content is crucial for comprehending the flow of information and the impact of digital surveillance in China. This paper characterizes these users across dimensions of geography, gender, social influence, and technology usage.
Methodology
The study utilized a web crawler to collect publicly available profile data and repost histories from Weibo. Censored content was identified by tracking the status of original posts within a user's repost chain; posts replaced by standard system deletion messages were flagged as censored. Users were categorized into two groups: the Censored group (those who reposted at least one eventually censored post) and the Uncensored group (those with no censored reposts). Statistical analyses, including proportional z-tests, were conducted to compare the distributions of features between these groups.
Social Capital and the "Caution of Influence"
A primary finding of this research is that social influence correlates negatively with the sharing of censored content. Users with high social capital—defined by large follower counts—and those with official certification (real-name verification) are significantly less likely to repost sensitive material.
As shown in the data, 86.3% of the Uncensored group had over 5,000 followers, compared to 80.2% of the Censored group. This suggests that users with established online reputations or those whose real identities are known to the platform exercise greater caution to avoid account suspension or other repercussions.
Demographic and Geographic Patterns
Geographic location serves as a significant predictor of a user's likelihood to share censored content. Users from economically vibrant regions, including Beijing, Shanghai, and the Greater Bay Area, were found to be more active in sharing sensitive posts. Notably, Hong Kong exhibited the highest rate of censored reposting (over 33%), likely reflecting its distinct political environment. Hubei province also emerged as a hotspot, attributed to the surge in expressive posting during the initial COVID-19 outbreak.
In contrast to geographic and social factors, gender does not appear to be a significant driver of censored sharing. The study found that males made up 44% of the Uncensored group and 46% of the Censored group, a difference that was not statistically significant.
Technological Proxies
Technological choices provide additional insight into user demographics. Apple device users were disproportionately represented in the Censored group. This may be a proxy for underlying demographics, as Apple products in China are often favored by younger, tech-savvy professionals who may be more inclined toward political discourse, whereas older users or those in the public sector may prefer other platforms or devices.
Conclusion
This study provides a demographic profile of the "censored user" on Weibo, highlighting that economic development, geographic context, and social capital are key determinants of online behavior under censorship. While influence breeds caution, specific regional events and socio-economic status can drive users to bypass self-censorship. Future research could further explore these dynamics by analyzing the comments associated with censored reposts to better understand the thematic nature of the shared content.
Summary updated