Comparative Analysis of News Audience Structures on Twitter and Weibo (MSc Thesis)

We hear a lot about how people’s opinions are supposedly shaped by their social news environments on platforms like Twitter (or Weibo, its Chinese counterpart). During my Master’s course at the Oxford Internet Institute, I wanted to turn this problem on its head. How do individual users contribute to, or enact, the contours of the broader news ecosystem when they share certain news articles and not others? In other words, rather than treating people as helpless and passive victims of online echo chambers, what if we looked at the social news environment itself as an emergent structure arising from the behavior of active agents?

To explore this problem, I turned to network analysis as a methodological approach, informed by theories of the “active audience” in the lineage of British cultural studies (Stuart Hall is one of my patron saints). Bringing in my background in Chinese language and prior work with Weibo, I put together a comparative research design that acknowledged and embraced the parochial nature of all social platforms—Twitter included—rather than pretending they could stand in for some imagined offline polity.

I was also interested in a few specific aspects of the news environments on Twitter and Weibo that seemed important and poorly understood: Internet censorship in China, political polarization in the US, and the effects of globalization. In addition to network analysis and cultural studies, I read up on comparative news systems, ethnographic perspectives on globalization and media use, and traditional audience analysis. 

I  brushed up on my Python skills to write a script that could crawl and parse raw HTML from Weibo keyword search results. Getting Twitter data was much more straightforward, since I could stream directly from the public API. I used Python and R to wrangle this data into a format that could be imported into Gephi, the open source network analysis software.

The specific methodological technique I used involved calculating the amount of overlap between the audiences of any two given news sources. By doing this for every possible pair, I could generate an intricate network map showing the way audiences actively construct the news landscape through acts of social sharing.

I constructed four such maps, comparing news sources on Twitter and Weibo by level of audience overlap for two sets of topics, roughly corresponding to “domestic” vs. “international” news on both respective platforms.

Twitter: US current events (left) and Chinese current events (right)
Weibo: Chinese current events (left) and US current events (right)

I found that when audiences are discussing more culturally proximate news topics—that is, when Twitter users talk about the current U.S. president, and when Weibo users talk about local Chinese issues—they are much more likely to splinter off into distinct subgroups. (Interestingly, while the Twitter network map was dominated by clusters corresponding to the political “left” and “right,” the Weibo map was much more diverse and heterogeneous.)

The opposite is true when more culturally distant news topics are discussed—when Twitter users talk about China, and when Weibo users talk about Donald Trump. In contrast with common perceptions about globalization, I interpreted this as evidence of the continuing importance of cultural homophily in structuring audience behavior (and of audience behavior in contributing to cultural homophily).

Check out my full thesis here.