An Efficient Fake News Detection System Using Contextualized Embeddings and Recurrent Neural Network

Fake news is detrimental for society and individuals. Since the information dissipation through online media is too quick, an efficient system is needed to detect and counter the propagation of fake news on social media. Many studies have been performed in last few years to detect fake news on social media. This study focuses on efficient detection of fake news on social media, through a Natural Language Processing based approach, using deep learning. For the detection of fake news, textual data have been analyzed in unidirectional way using sequential neural networks, or in bi-directional way using transformer architectures like Bidirectional Encoder Representations from Transformers (BERT). This paper proposes Contextualized Fake News Detection System (ConFaDe) a deep learning based fake news detection system that utilizes contextual embeddings generated from a transformer-based model. The model uses Masked Language Modelling and Replaced Token Detection in its pre-training to capture contextual and semantic information in the text. The proposed system outperforms the previously set benchmarks for fake news detection; including state-of-the-art approaches on a real-world fake news dataset, when evaluated using a set of standard performance metrics with an accuracy of 99.9 % and F1 macro of 99.9%. In contrast to the existing state-of-the-art model, the proposed system uses 90 percent less network parameters and is 75 percent lesser in size. Consequently, ConFaDe requires fewer hardware resources and less training time, and yet outperforms the existing fake news detection techniques, a step forward in the direction of Green Artificial Intelligence.