Amina Yusuf Umar, Ibrahim Said Ahmad, Khadijah Muhammad

Keywords: Deep learning, Fake news detection, Misinformation, Disinformation, Rumors, CNN, GRU.

Abstract

Social media platforms enable the creation and sharing of news, information, ideas, and other expressions. However, it also facilitates the dissemination of fake news and misinformation. Therefore, there is a need for effective methods of detecting fake news on social media platforms. Consequently, fake news detection has attracted many studies in recent years. Notwithstanding, there exist various flaws due to the characteristics of the algorithms used and the nature of the data. This work addresses some of the limitations of a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) approach in fake news detection. Specifically, large computational complexity and storage of a large number of redundant intermediate variables of the LSTM algorithm. The limitations are addressed by proposing a hybrid of CNN and Gated Recurrent Unit (GRU) instead. GRU is similar to the LSTM, but with fewer computational complexities. The proposed approach was evaluated using four existing datasets based on accuracy, precision, and recall. The result of the evaluation shows the superiority of the proposed approach over the baseline approach.