Cross-Platform Information Flow and Multilingual Text Analysis

A Comparative Study of Weibo and Twitter Through Deep Learning

Abstract

This study delved into cross-platform information flow and multilingual text analysis by examining social media posts on Weibo and Twitter in Chinese and English. We investigated public opinions about a violent restaurant attack in China that received widespread attention and validated three strategies of Bidirectional Encoder Representations from Transformers (BERT) to classify multilingual social media posts regarding their attitudes, targets, and frames. This study found that there was more criticism than support on Twitter than on Weibo when calling for social justice. When targeting the governments, Weibo users focused more on the local level, while Twitter users focused more on the state level. When framing their opinions, Weibo users focused more on gender violence, while Twitter users focused more on gang violence. These variations within social media posts across platforms were fundamentally influenced by the interruption of transnational information flow as a result of Chinese governance and censorship of the internet. Through the “porous censorship,” social media users’ autonomy and trust in the government played critical roles in the dynamics between online criticism and authoritarian responsiveness.