COVID-19 Fake News Detection in Malaysia – A Supervised Approach

Abstract

Social media has been flooded with enormous amounts of COVID-19-related information ever since the COVID- 19 pandemic started back in 2020. Since then, Malaysian citizens have become more reliant than ever on social media for consumption of COVID-19 information. However, the lack of COVID-19 news regulations on social media platforms encouraged people to post unverified, fake and misleading COVID-19 related information. Because of the time-consuming nature of fact-checking, people often take these unverified COVID-19 news for granted. Consequently, people inadvertently spread these fake COVID-19 news to their families, friends and relatives on social messaging platforms like WhatsApp. The spread of COVID- 19 fake news online in Malaysia can have severe sequences, causing widespread panic among fellow Malaysians. In this paper, we proposed a supervised learning approach to detect COVID-19 fake news. The fake news on COVID-19 were scraped from the website called Sebenarnya, and real news were scraped from The Star website. We applied a semantic model with different word representations which include Bag of Words (BOW), Term Frequency - Inverse Document Frequency (TF-IDF), Word2Vec and Global Vectors (GloVe). In the evaluation step, 6 supervised machine learning algorithms were applied such as Multinomial Naive Bayes, Support Vector Machines, Decision Tree, Random Forest, Logistic Regression and Adaboost. Afterward, 10-fold cross validation was used to train and evaluate the 6 supervised algorithms according to performance metrics such as accuracy, precision, recall, AUC-ROC, F1-score. The results showed that Random Forest with the word representation of TF-IDF per- formed the best with over 97% accuracy in contrast to numerous conventional supervised machine learning classifiers.