Hybrid-Multi-channel Deep Neural Network for Fake News Detection

Nowadays, the peril of fake news is badly affecting every individual, from a common man to the president of the world’s top country. In this Internet era, fake news spreading on social media is getting easier day by day. Initially, traditional machine-learning algorithms were used to identify fake news. With the advent of many deep-learning techniques, significant research work is being carried out on the application of deep learning techniques in fake news detection in recent years. In this paper, we propose a hybrid-multi-channel deep neural network (MCDNN) model with word embedding (Glove) for content-based fake news detection. This paper demonstrates the better performance of the proposed model compared to the other standard deep learning models; a simple multi-layer perceptron (MLP) neural network, long short-term memory (LSTM), combined LSTM plus CNN (Convolution Neural Network) network, and Gated Recurrent Unit (GRU) with hyper parameters tuned for optimal performance. The proposed hybrid-MCDNN model was observed to give 99.60% accuracy of fake news detection on news collected at the 2016 US Presidential election time and 83.53% on the LIAR dataset.