oa Structural And Semantic Evolution Of Egyptian Political Polarization On Twitter

At this point there is little question about whether online social networks have occupied (and occupy) a central spot as the backbone of modern civil protests. This agreement has triggered a whole research avenue, in which social structure (who is friends with whom) and content analysis (what is said online) have been central -though generally approached separately. Countries in the MENA region are prominent examples of turmoil online case studies: Iran (2009), Tunisia, Lybia, Egypt and Bahrain (2011), ongoing Gaza conflict and Syria (2011-present), have witnessed social revolts, violence and even some government changes. Focusing on Egypt, its society is known to be highly polarized with two dominant poles typically labeled as "Islamist" and "Secular". Since the coup/revolution in July 2013, a new pro-military intervention vs. anti-military intervention (pro-MI/anti-MI hereafter) dimension has emerged, though it is widely considered in mass media to be aligned with the previous polarization in the sense that Islamist/Secular coincide with anti-MI/pro-MI. However, during the violent aftermath of the army's ascent to power some observers speculated that certain parts of the pro-MI supporters might have changed their opinion due to what some view as excessive use of force against protesters, with at least several hundred killed. In this work, we examine how these events have transpired on Twitter, exploiting a set of nearly 6 million Arabic tweets authored by over 120,000 users. We question whether opinion changes are due to (i) people actually switching sides, or (ii) the relevant camps becoming increasingly louder or more muted. In addressing these questions, we learn about the underlying psychological and sociological mechanisms which are at play in conflictive contexts. For instance, it might be more benefitial -or less costly- for an individual to withdraw from a conflict than to actually switch sides: with a similarly minded social neighborhood, an actor is faced with the ``volunteer's dilemma'' (who switches side first); from which the rational outcome is to become mute. We approach these questions both from the structural (network) side and the semantic (content) side. Concerning content, we used a set of manually labeled hashtags to build a pro- or anti-MI classifier, which is used to classify tweets based on its textual content. On the network side, we used retweets of hand-labeled seed users to derive a Secular vs. Islamist leaning for users. We find that despite -or because of- the dramatic events there is little evidence of users changing sides. We look at switching between Secularist and Islamist camps and between pro-MI and anti-MI camps. Our network and content analyses indicate that less than 5% of users switched sides. Instead, the narrative seems to be one of pro-MI and Secular users being dominant in terms of volume leading up to July 3, and anti-MI and Islamist users gaining in volume afterwards. Furthermore, in contradiction to the dominating narrative in mass media, the correlation between being a secular and pro-MI is far from perfect. However, some correlation was noticed between being an Islamist and against the military intervention.