Efficient Machine Learning Approaches to Detect Fake News of Covid-19

Abstract

In recent years, the growth of fake news has been significantly high. Advancement in the field of technology is one of the reasons that lie behind this phenomenon. Fake news are presented in such a way that it is quite hard to identify as fake on various social platforms these days and that has a huge impact on people or communities. Such fake news is most destructive when it plays with life. COVID-19 has changed and shaken the entire universe, and fake news that are related to COVID-19 make the destruction deadlier. So, an effort regarding COVID-19-related fake news detection will guard a lot of people or communities against bogus news and can make lives better with proper news in a pandemic. For our research in this paper, a methodology has been espoused to detect COVID-19-allied fake news. Our methodology consists of two different approaches. One approach deals with machine learning models (Logistic regression, support vector machine, decision tree, random forest) using the term frequency-inverse document frequency (TF-IDF) attributes of textual documents, and the other approach involves an association of convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) using the sequence of vectors of tokens or words in documents. Logistic regression using the TF-IDF is the best performer among all these models having 95% accuracy and an F1-score of 0.94 on test data with Cohen’s kappa coefficient of 0.89 and Mathews correlation coefficient of 0.89.