Stacked Bi-LSTM with Attention and Contextual BERT Embeddings for Fake News Analysis

With the evolution of digital technology and media, there has been a surge in the number of internet users worldwide. Due to this growth, it has become necessary to stop fake news from being shared on social media platforms as it can mislead large sections of society by spreading unverified information. Therefore, it becomes vital to create a framework for the automatic verification of content shared online. In this work, we propose a deep learning framework using stacked Bi-LSTM with self-attention to detect fake news articles. The proposed model uses contextual embeddings pooled from BERT large model by fine-tuning its last ten layers on the training dataset. The proposed framework's effectiveness is being tested on four different open-source datasets, namely, News Articles Dataset, Kaggle Dataset, Gossipcop Dataset, and Politifact Dataset, through extensive experimentation and evaluation. The highest accuracy achieved by the proposed approach is 0.99 on the News Articles Dataset and Kaggle Dataset. State of the art comparison with contemporary methods shows that our proposed method achieves better performance than peers.