Divide et impera : how cyberbalkanized social media tear apart our society

Abstract

In most of the developed countries, more than half of the population obtain political information from social media such as Facebook. Tons of evidence suggest that Facebook users tend to interact only with friends and information sources with political ideology similar to their own. This phenomenon, namely cyberbalkanization, has been observed globally but is poorly studied. Its generative mechanism and correlates with other online behaviors are yet to be explored. Despite widespread public concern over the claim that cyberbalkanization polarizes public opinion, the relationship has not been tested empirically. This dissertation attempts to examine this assertion and explore its mechanism. In order to approach this problem, a two-level hierarchical conceptualization of cyberbalkanization is proposed: Facebook Page-level cyberbalkanization describes page sharing contents mainly from pages with similar political inclination and Facebook content consumers level cyberbalkanization represents users seeking contents preferably from like-minded pages. Social media data were collected during the 2014 Occupy Movement in Hong Kong. A post-sharing network was generated by communication among 1,644 Hong Kong Facebook Pages. The users who engaged with the sampled pages, such as likers and commenters, were also obtained. A multi-method approach was used to analyze these data. Study 1 analyzes the cyberbalkanization of the page network and to test a generative mechanism based on social balance theory. Clusters of pages with clear political leanings were extracted. Using agent-based modeling approach, the formation of ‘enemies of friends are my enemies’ triad relationship was found as a necessary condition to the formation of these clusters. Study 2 investigates the association between page- and user-level cyberbalkanization. Using network regression approach, we found that two pages share each other’s posts more frequently have more common users. Therefore, page- and user-level cyberbalkanization are intertwined. Study 3 examines the content shared within and between clusters of pages and to test the possible ‘nasty effect’ (the polarization effect of incivility online content) of cyberbalkanization. A representative sample of shared uncivil contents were manually coded. This analysis revealed uncivil content was significantly more likely to be shared within political clusters but not between them. Study 4 explores the temporal relationship between page cyberbalkanization and offline opinion polarization using time series analysis. The analysis found the change in degree of cyberbalkanization was temporally preceded the change in telephone poll’s opinion polarization for over 20 days. This dissertation substantiates cyberbalkanization may induce opinion polarization via a self-organized ‘divide et impera’ (‘divide and conquer’) mechanism, i.e. page cyberbalkanization segregates users apart (‘divide’ ); users are then further administered by issue-framing using uncivil content by the pages shared within isolated political clusters (‘conquer’). Understanding this mechanism is helpful to prescribe remedies to mitigate the potential threat of cyberbalkanization to deliberative democracy. Some possible remedies are suggested.