A fusion of BERT, machine learning and manual approach for fake news detection

A large number of users around the globe have preferred to read news and the latest information from the Internet, especially social media, leaving behind the traditional approach of print media. On the one hand, the Internet is a constructive medium to spread the latest news and information briefly. On the other hand, malicious users are very active on the Internet and spread fake news, which becomes viral within a few minutes. The spread of fake news has become a serious threat as many users now rely on Internet news without verification. In this digital world, it is easy to spread any toxic information over the Internet, like hate speech, extremism, propaganda, and political agendas. It is a big challenge in today's digital world to mitigate the spread of fake news; hence, there is a need for an automatic computational tool that can assist in measuring the credibility of news. This study aims to deliver a solution where fake news from Twitter and website-based articles can be detected using the Natural Language Processing (NLP) technique, Bidirectional Encoder Representations from Transformers (BERT), other machine learning classification algorithms, and manual program-based approaches. A dataset with fake and real labels for the textual content is used. Different classification algorithms are evaluated to find a suitable algorithm for delivering a fake news detector. The evaluations are based on machine learning and a program-based approach. The textual content that the user provides, such as an article or tweet, can confirm the legitimacy of fake news. This website offers fake news detection for both website-based news articles and tweets from Twitter in English, Arabic, and Urdu.