Fake News Detection in Mainstream Media Using BERT

Abstract

AbstractFake news classification is undoubtedly a challenging task that demands to be resolved in this world of media and mass communications. Despite the majority of the falsified information being strictly filtered, some of the fake news makes it into platforms of large audiences such as news and radio channels. This paper proposes a relatively simple yet reliable solution for real-time fake news classification focused on mass media sources such as news and radio channels. The proposed model focuses on the user’s multiple ways to input the news from live media from which relevant data can be extracted. Then, it utilizes a natural language processing (NLP) framework called BERT to extract contextual features from the raw text data and classify the news as fake or real. Experimental results obtained reflect the effectiveness of the model in detection of fake news when the headline and supporting text are provided as an input. The proposed model achieves an accuracy of 96.5% on the test set classification into two classes—real and fake.KeywordsFake newsMachine learningNatural language processingBERT