ScrutNet: a deep ensemble network for detecting fake news in online text

Abstract

The expeditious propagation of fake news through online social media platforms has cropped up as a captious challenge, undermining the credibility of information sources and affecting public trust. Accurate detection of fake news is imperative to maintain the integrity of online content but is constrained by availability of data. This research aims to detect fake news from online articles by proposing a novel deep learning ensemble network capable of effectively discerning between genuine and fabricated news articles using limited data. We introduce ScrutNet, which leverages the synergistic capabilities of a bidirectional long short-term memory network and a convolutional neural network, which have been meticulously designed and fine-tuned for the task by us. This comprehensive ensemble classifier captures both sequential dependencies and local patterns within the textual data without requiring very large datasets like transformer based models. Through rigorous experimentation, we optimise the individual model parameters and ensemble strategy. The experimental results showcase the remarkable efficacy of ScrutNet in the detection of fake news, with an outstanding precision of 99.56%, 99.43% specificity, and an F1 score of 99.49% achieved on the partition test of the data set. Comparative analysis against state-of-the-art baselines demonstrates the superior performance of ScrutNet, establishing its prominence as a generalised and dependable fake news detection mechanism.