Enhancing Fake Review Detection Using BERT Transfer Learning Algorithm in Natural Language Processing

– In recent years one of the biggest world concerns is the proliferation of fake news through social networks. Due to their intended purpose as means to sway the beliefs of large crowds, fake news has been causing a great effect on the world. Much attention has been paid to this area by researchers as the process of manual confirmation of the news’ authenticity is practically impossible and very costly. Explorations into the detection of false news targeted content based approaches, social context based approaches, image based approaches, sentiment based approaches and hybrid context based classification systems. As a consequence of utilizing the content-based classification approach, this work will propose a model for False news Classification utilizing the headlines of the news. The model used to solve the current problem is a BERT model and the output layer of which is connected to an LSTM layer. The FakeNewsNet dataset was used in both the training and evaluation processes, which consists of two sub-datasets: PolitiFact and GossipCop. Comparison has been made between the model and basic classification model. The suggested model, which utilizes an LSTM layer for the evaluation of impact, is almost analogous to the vanilla BERT model that was trained on the dataset on the same terms and conditions. The findings showed that there was a 2 percent increase in the level of accuracy that has been achieved. For PolitiFact, the recall was equal to 45 % and for training a 1. The performance boost achieved on the GossipCop dataset is about 11% as compared to the vanilla pre-trained BERT model.