SentiMage: A Sentiment-Image-based COVID-19 Health Misinformation Detection using Machine Learning

The rapid dissemination of misinformation (generally known as fake news) has become worrisome, especially during the on-going COVID-19 pandemic both globally, and locally. In fact, the proliferation of health-related misinformation intensified on social media, which many experts believe is contributing to the threats of the pandemic. Sentiment has been shown to improve detection mechanisms in various social media related studies, however this aspect is under-researched in the context of health misinformation. Further, metadata such as location or image that constitute part of real and fake news were not fully explored as well. This study develops a health misinformation detection model using machine learning algorithms, and further assesses the impact of sentiment and image on the model performance. Local data gathered from a fact-checking portal were pre-processed, translated, and used to train the detection model. Evaluation results show Support Vector Machine to yield the best performance with 99.4% for F-measure and accuracy of 99.1%, followed closely by Random Forest when sentiment was included, however, the presence of image was not found to significantly improve health misinformation detection.