Enhanced Fake News Detection Leveraging a Hybrid BERT-BiLSTM Model

Fake news is phenomena that deserves global attention especially with the above issues of social media being the largest source that influences people’s opinions, attitudes or decisions. Authenticating news in the upswing is energy-consuming and highly priced which is why there has been a growing focus in the field. Different means of solutions in issues related to the faking of news has been sought including: content-based classification, contexts based classification, images based classification, sentiment analysis, and mixed approaches. This work offers a method of fake news detection where news headlines are considered, which falls under the content-based classification scheme. The Architecture incorporates BERT and encompasses an LSTM layer as well. Training and evaluation were done on FakeNewsNet - PolitiFact , GossipCop datasets and ISOT dataset. The method propounded was tested against benchmark classification models while the impact of the LSTM layer on classification was tested by a BERT model that was subjected to the same conditions.