Comparative Analysis of Transfer Learning and Attention-driven Memory-based Learning for COVID-19 Fake News Detection

Abstract

In the pandemic COVID-19 situation, the world is facing a pandemic of fake information which often stirs the public attention by attacking their emotional quotient. Scenario reached a situation where people in search of worthy information for public health and precaution, getting fake news. This unprecedented expansion of fake information has become a challenging research issue. Deliberate efforts have been attempted in this manuscript for finding a solution to this COVID-19 fake news detection problem with the help of deep learning models. Two deep learning models—BERT, a transfer learning model, and attention-based bi-directional long short-term memory (LSTM), a memory-based model, have been applied in order to get accurate fake news classification outcomes. A comparative outcome of both models is presented which shows BERT outperforms and gives excellent results in comparison to the attention-based bi-directional LSTM model. The achieved training accuracy by BERT is 86% which is much higher than the accuracy achieved by attention-based Bi-LSTM. BERT precision, recall, and F-score are 0.82, 0.79, and 0.80, respectively, which shows that BERT can detect COVID-19 fake news better than the attention-based Bi-LSTM model.