A longitudinal sentiment analysis of the #FeesMustFall campaign on Twitter

Abstract

The #FeesMustFall campaign began in 2015 to lobby government to provide students with free university education in order to redress past imbalances. It rapidly progressed to become a widespread national phenomenon that attracted international attention and sympathetic support. However, certain unsavoury incidents marred the campaign and attempted to derail it from achieving its goals. The campaign did reach many of its targets with the South African government eventually announcing free education for the poor and working class in December 2017. #FeesMustFall has been well documented and researched, however, no literature offered a quantitative insight into the opinions of social media users during this campaign, although a unique feature of #FeesMustFall was leveraging social media platforms to coordinate the campaign. This study addresses this gap by undertaking a longitudinal sentiment analysis of textual conversations expressed on the Twitter social media platform. This longitudinal study analyses the Twitter #FeesMustFall campaign through the acquisition of 576 583 tweets posted between 15 October 2015 and 10 April 2017. These tweets were pre-processed and cleaned by removing exact duplicates and unintelligible data. The research method to analyse the “cleaned” #FeesMustFall data utilises, inter alia, descriptive statistics, sentiment analysis using a natural language programming (NLP) approach called Valence Aware Dictionary sEntiment Reasoner (VADER) and code written in Python. VADER is a lexicon rule-based sentiment analysis tool particularly suited to social media. To detect multiple changes in this large historical dataset, the Change Point Analysis method (CPA) is applied using a Cumulative Sum Analysis (CUSUM) method to identify changes across time. The research question is whether and for what reason the online sentiment changed during the observation period. The sentiment expressed is triangulated with perceived real-life negative events, such as the burning of the University of KwaZulu-Natal (UKZN) library and the University of Johannesburg (UJ) Hall, to understand whether online activism sentiment reflected or reacted to real-life events. The study finds that sentiment did change in relation to these two events, one on the day of the UKZN library event and one prior to the UJ Hall event. Social robots (bots) are automatic or semi-automatic computer programs that mimic human behaviour in online social networks. Their deployment exposes online activism to manipulation. A further research question addressed whether bots played a role in the #FeesMustFall campaign. A review of bots, their characteristics, behaviour, and detection methods was undertaken. The study does indeed establish the presence of bots during #FeesMustFall. The study’s contribution is significant as this is the first longitudinal study of the #FeesMustFall campaign which observes the sentiment distribution and changes. It is also the first study to investigate and find evidence of bots in the #FeesMustFall campaign.