Automated Fake News Detection for societal benefit using Hybrid Deep Neural Network

Abstract

In recent years, the emergence of Online Social Networks has led to a profusion of social news such as commercial advertisements, political news, information about celebrities, and many other types of information. Users of social media platforms like Facebook, Instagram, and Twitter have been influenced by false information. The proliferation of false news across several industries and government bodies is a direct result of the rise of social media and online discussion forums. As a result, people are less likely to put their faith in the media. There is a mountain of literature on the topic of Artificial Intelligence (AI) methods for spotting hoaxes. The aforementioned problem is solved by employing a hybrid Neural Network design that incorporates the strengths of both CNN and LSTM. Dimensionality reduction strategies were recommended for use in this study to help make feature vectors more manageable before being sent on to a classifier. To do this, we construct a multi-layered Convolutional Neural Network (CNN). We evaluate the suggested method by contrasting it to many standard templates. The proposed model was trained and tested using state-of-the-art benchmark datasets, with test-data accuracy of 98.36%. The results were verified using a battery of performance assessment metrics including false positives, true negatives, precision, recall, F1, accuracy, and more. These results reveal considerable gains in the field of fake news identification as compared to previous state-of-the-art results and validate the promise of our technique for identifying false news on social media. This study will help academics gain a deeper comprehension of how hybrid CNN-based deep models may be used to spot false news.