B-LIAR: A novel model for handling Multiclass Fake News data utilizing a Transformer Encoder Stack-based architecture

In today's digital era, social media is involved in purposely pushing fake news over the web to increase the readership of the people. Various techniques that are adopted for propagation are clickbait that works by posting content that attracts the users with flashy headlines to increase their advertisement reviews. Other methodologies include satire content which publishes fake news mainly for entertainment, sloppy journalism where sometimes reporters or journalists publish stories without complete reliable information which leads to misinformation to the audience. Therefore, a detection system is necessary as it can help national security agencies, investigation departments, businesses, the industry in detecting fake news in social media. In this paper, for the detection of fake news, we have used the LIAR dataset. This dataset consists of 12.8k manually labelled sentences from PolitiFact.com in various contexts. We have deployed a novel attention approach for detection using BERT embedding in our proposed model. Using our proposed model, we have reached the state-of-the-art result compared to the existing methods. The results obtained from our experiments showcase that our model can be used by researchers for further exploration.