Protest events have a 'Twitter Signature': Evidence from South Africa's #FeesMustFall

Abstract

Protest levels in South Africa remain notoriously high. However, Protest Event Analysis (PEA) within the South African context has predominantly relied on media or administrative data, with scant regard for the insights that social media records may provide. Using data gathered during the #FeesMustFall movement, we employ a machine learning model trained on historical South African Twitter data combined with event records from a protest database. Our findings indicate that such events establish a distinctive time-series pattern, suggesting a robust approach to modelling, identifying or predicting protests within South Africa. Moreover, this implies that protests can be characterised using social media sources in ways that can enhance the insights from traditional data sources, and that automated analysis can unearth and assemble PEA databases. Moreover, automated PEA may soon provide valuable information about the actors, level of turmoil, size, duration, grievances, and motivations of protest.