On the Effect of Intelligent Neurally Processed Features on COVID–19 Fake News Detection Performance

Abstract

Online social networks are flooded with lot of user-generated information; fake news offenders use these online social network platforms to spread COVID fake news. This propagation of fake news results in a low level of content truthfulness, distrust in online social networks, panic and fear, which makes people take erroneous decisions. Hence, accurate classification of fake news against real news is mandatory. Therefore, in this study, a novel Neurally Augmented model is proposed to classify fake news accurately based on the content of the tweet. The proposed model creates a novel Deep Neural Network to automatically extract the neutrally processed features for downstream processing in a classic Machine Learning model. A neurally processed feature from 1D ConvNet is used to augment classic Machine Learning model in an attempt to improve the classification and discriminative capability of the classic Machine Learning model. Experimentation is done on a COVID-19 Twitter dataset curated by the authors. The proposed methodology provides a highly accurate fake news classification of 97.25%, which is 12% superior to the classic Machine Learning and Deep Learning models without neural augmentation. The proposed methodology is further evaluated on the ISOT, Indian Fake News Dataset, LIAR and Constraint Shared task COVID-19 Fake News benchmark datasets to determine its robustness, and it achieves significant accuracy of 93.75%, 89.17%, 83.68%, and 91%, respectively. The proposed model is tested on the proposed dataset and achieved a 5% increase in accuracy over the benchmark datasets.