Hybrid Neural Network Models for Detecting Fake News Articles

Abstract

Abstract The prevalence of world-wide access to the Internet has come at a cost. A lot of misleading information is posted on public news websites and social media. Many news writers and organizations manipulate their posted data to propagate false information that target different societies and in different languages. Accurate and timely detection of false news is made possible in large part using machine learning-based technologies. This paper targets the problem of detecting fake news in Arabic language using machine learning models. A hybrid model of two deep neural networks is used to classify Arabic news articles in order to detect fake articles. The two types of neural networks are convolutional and bi-directional long-short term memory. Robust features are extracted using two different word vectors and a complex model of a convolutional neural network. Moreover, a set of auxiliary output layers are used to enhance the model accuracy. Multi-class classification is achieved via modifying the primary output layer. Results show an accuracy of 88% and 78% for binary classification and multi-class classification, respectively.

Keywords

• Fake news
• Neural networks
• Multi-class classification
• Binary classification