Arabic fake news detection for Covid-19 using deep learning and machine learning

Abstract

When newspapers were the dominant form of conventional media, fake news was widespread. Due to the vast influence of such false news and the growing user reach of technical media sources (TV, Internet, social media, blogs). Humans have become more dependent on the news as they make daily decisions for ensuring the safety of their loved ones and themselves in the wake of COVID-19 becoming a pandemic which has impacted humans all over the world. Fake news, on the other hand, is on the verge of becoming a "second pandemic" or "infodemic," endangering the health of individuals all over the world. Previous research hasn't used fake news detection to coronavirus in Arabic due to the fact that fake news connected to coronavirus is such a recent occurrence. A total of 4 versions of the datasets used in this study have been produced (D0, D1, D2, and D3). To understand the effects of deep learning (DL) and machine learning (ML) techniques on any dataset, a total of 4 datasets were created. Also, the research analyzes them with regard to ML and DL to determine the efficacy of preprocessing (D1), raw dataset (D0), light stemming (D3), and root stemming (D2). Dataset version zero (D0) is finished when creating an excel file. From the first version (D0), three more versions (D2, D1, and D3) were created. This study examines the detection of fake news articles concerning COVID-19 on Facebook with the use of DL approaches, like the Bidirectional Long Short-Term Memory Networks (Bi-LSTM), Bidirectional Encoder Representations from Transformers (BERT) and AraBert of Arabic text and ML techniques Linear Support Vector Machines (SVM) and Random Forest (RF). On testing data-set (D0), BERT yields the greatest accuracy of 97.32%