A Computational Analysis of News Media Bias: A South African Case Study

Abstract

News media in South Africa is assumed to be unbiased and objective in their reporting of the news. Indeed, editors are required to uphold an objective and balanced view with no favour to external political or corporate interests. This assumption of objectivity is tested on a large scale by computationally analysing 30 000 articles published by five media houses: News24, SABC, EWN, ENCA, and IOL. Using topic modelling, 38 topics are extracted from the corpus, and sentiment is computed for each topic. The study highlights various cases of both over and under-reporting by media houses on particular topics. We also identify various tonality biases by media houses.