A Discriminative AI-Based Sequential Classification Model for Fake News Detection using BERT

Abstract

In today’s world, there is a widespread flow of data and information due to immense usage of web applications. The usage of information available on web is difficult to be categorised and identified as true or Fake for a normal user. This creates a sort of doubt and confusion in the mind of normal users as how to verify and authenticate the news in the social media. This is essential in today’s modern-day scenario as regulatory bodies do wants and enforce that the true news or information should be further circulated or propagated rather than fake news which at times is not known to common user and he may be mis leaded by fake news. Public confidence in widely disseminated news has been undermined by the ease with which disinformation may now reach large audiences due to immense growth of social media and online platforms. To overcome such situation, we need a well-defined system which ensure the identification of fake and true news with highest accuracy. The authors proposed a text-based sequential classification model for detecting fake news using Bidirectional Encoder Representations from Transformers (BERT). The model proposed is able to clearly classify the news as fake or true with enhanced accuracy and precision. The model will be able to accurately identify the Fake news and will benefit the society by helping to clearly distinguish between fake and true news. The model is built on usage of transformer model which has ability to learn from complex relationship in label data and enhances the performance of predicted model with faster training time due to parallel processing.