Mining tweets on sexual violence in South Africa

Abstract

The “wisdom of the crowds” emerges from mining tweets incited by the behaviour of the masses. South Africans have resorted to venting their frustrations regarding the state of Gender-based violence (GBV) on social media. This is done in a bid to seek justice and raise awareness of the issue at large. Sexual violence is a kind of GBV that is widespread in South Africa, and Twitter is that popular social media and micro-blogging service where these frustrations are mostly reported as tweets. An effective analysis of Twitter data on sexual violence can expose unknown information and provide insights leading to better mitigating strategies. Within this context, this research investigates the disparity in reported cases of sexual violence in terms of gender representation and sentiment expression based on geolocation. It is envisaged that this information can be used in the form of an interactive data visualisation tool that policy analysts can use to investigate counterpreventive measures. Moreover, law enforcement agencies can better allocate their resources to help mitigate this problem. To do this, we achieved the development of a web-based, AI-driven application that automatically reveals the gender and sentiments in a tweet document. The system is built using the Estimator framework, a TensorFlow high-level-API which simplifies machine learning programming and model development. The process is initiated by the Indexer which is a server-side Node.js application that runs persistently and enables streaming of Twitter data for gender prediction, sentiment analysis and visualisation of GBV in South Africa