https://orcid.org/0000-0003-1674-1543
![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
ABSTRACT
The rise of the ‘alt-right’ (alternative right) and their communications on the Internet are not unique to the West. This study follows a mixed-methods approach combining topic modeling, social network analysis, and discourse analysis to analyze the discursive and network structure of an online Chinese alt-right community on Weibo. We summarize the topics Chinese alt-right influencers discuss and examine how these topics are interrelated. We find that the Chinese alt-right discourse can be deemed as both an extension and localization of the global alt-right: they frequently discuss global alt-right issues and also hold alt-right ideologies on domestic issues. Meanwhile, influencers in the community are densely connected, suggesting a high level of coordination and cooperation. We particularly identify two discursive strategies that alt-right influencers employ to reproduce the transnational alt-right discourse, namely invented common crisis of majority culture and transnational metaphor usage. These findings provide insights into the transnational aspect of the rise of global alt-right.
Acknowledgements
The authors would like to thank anonymous reviewers and Guobin Yang, Daniel Hopkins, Yilang Peng, Sijia Yang, and Diami Virgilio for providing helpful feedback to the project. The Authors also want to thank Dr. Ruth Ben-Ghiat for her inspiring spring 2019 seminar on “Propaganda and Media in Democracies and Dictatorships”, which was sponsored by the Center for Media at Risk at the Annenberg School for Communication.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
2 See https://www.jiguang.cn/reports/381 (in Chinese).
3 These keywords include ‘feminist bitch’ (女权婊), ‘feminist fist’ (女拳), ‘feminist dog’ (女犬), ‘feminist’ (女权分子), ‘pastoral feminism’ (田园女权), ‘Black Lives Are Expensive’ (黑命贵), ‘n****r’ (黑鬼), ‘halal’ (清真), ‘White Left’ (白左), and ‘Leftard’ (左棍), which assign negative connotations to feminists, black people, ethnic minorities, Muslims, and liberals who support the marginalized groups.
4 We used our definition of ‘alt-right,’ discussed in the literature section, to determine whether a post qualified as alt-right (whether it rejects the value of others). Due to the limited number of accounts we searched, we did not follow a conventional content analysis procedure but used a strict standard: only accounts that both researchers agreed to include were further studied.
5 We did not use Weibo API because it imposed restrictions on data collection. The Selenium package mimicked browsing behaviors of ordinary users. In this study, posts we gathered were public, accessible to any ordinary Weibo user.
6 A post originating from Account A reposted by Account B could also appear on the timeline of Account C who further reposts from Account B (C → B → A). The post on the timeline of C could be decomposed into two reposting links (C → B and B → A) and three segments of post (original post by A, repost by B, and repost by C).
7 While posts engaged by only one influencer might also be alt-right posts, we believe those with two or more influencers engaged should better represent the alt-right discourse in this community.
8 As an exploratory method which summarizes the topics frequently appearing in texts, LDA is a common topic model widely used by computational social scientists. The model generates each topic with a list of words and their coefficients associated with the topic so that this structure of topics conforms to certain statistical distributions. After creating the model, we can calculate the associations of each text with each topic.
9 According to the analyses of topic modeling, Topics A, B, and C were categorized as domestic topics and Topics D, E, and F were categorized as global ones, which met the face values of two clusters. We also conducted a test to further support this categorization: topics within the same group should have a higher level of similarity, which can be characterized by the measure of topical cosine similarity. By generating 1,000 subsets of bootstrapped samples from the overall posts, we found that the average cosine similarities between topics in the same topical cluster were significantly higher than the cosine similarities of two topics from different clusters (similarities among local topics vs. similarities between local and global topics: p < .001; similarities among global topics vs. similarities between local and global topics: p < .001). This result further provides confidence in our topic models and categorization (see ).
10 A weakly connected component is a subgraph of the overall graph within which any two nodes have a path between them, ignoring the directions of edges. A strongly connected component is a subgraph of the overall graph within which any two nodes can reach each other, considering the directions of edges.
11 SHIFTlocalTOglobal and SHIFTglobalTOlocal were measures we used to identify topical transitions (from a local topic to a global one, or from a global topic to a local one), which were built upon results of the topic model. The number here was associated with one segment of the post with a significant topical transition. The whole post including that segment was selected for qualitative analysis. For more details regarding these measures, see Appendix C.
Additional information
Notes on contributors
Tian Yang
Tian Yang is a PhD candidate at the Annenberg School for Communication, University of Pennsylvania. His research interest is at the intersection among political communication, computational social science, and social networks.
Kecheng Fang
Kecheng Fang is an Assistant Professor at the School of Journalism and Communication, The Chinese University of Hong Kong. His research interests include digital media, journalism, and political communication [email: [email protected]].