Development of a WhatsApp Bot to Detect Disinformation

Abstract

Disinformation poses a significant threat to democracy in South Africa, particularly due to its rapid distribution through social media platforms. Existing disinformation detection research largely overlooks African contexts. This research addresses this gap by developing a disinformation detection system for the South African context, which encourages a human-in-the-loop approach to disinformation. It incorporates state-of-the-art Natural Language Processing (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$N L P$</tex>) models, including Bidirectional Encoder Representations from Transformers (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$B E R T$</tex>) and Long ShortTerm Memory (LSTM). The system integrates these models with a WhatsApp bot interface that indicates disinformation probability and encourages further investigation. The BERT model achieved a remarkable accuracy of 99.40 %, while the LSTM model reached 88.91 %, demonstrating the system's potential to mitigate the spread of disinformation within the region significantly. Another key contribution of the work is the creation of a new, contextually relevant dataset labelled as either “realxy” or “disinformation,” comprised of South African news articles and Twitter data.