Emotions unveiled: detecting COVID-19 fake news on social media

Abstract

The COVID-19 pandemic has highlighted the pernicious effects of fake news, underscoring the critical need for researchers and practitioners to detect and mitigate its spread. In this paper, we examined the importance of detecting fake news and incorporated sentiment and emotional features to detect this type of news. Specifically, we compared the sentiments and emotions associated with fake and real news using a COVID-19 Twitter dataset with labeled categories. By utilizing different sentiment and emotion lexicons, we extracted sentiments categorized as positive, negative, and neutral and eight basic emotions, anticipation, anger, joy, sadness, surprise, fear, trust, and disgust. Our analysis revealed that fake news tends to elicit more negative emotions than real news. Therefore, we propose that negative emotions could serve as vital features in developing fake news detection models. To test this hypothesis, we compared the performance metrics of three machine learning models: random forest, support vector machine (SVM), and Naïve Bayes. We evaluated the models’ effectiveness with and without emotional features. Our results demonstrated that integrating emotional features into these models substantially improved the detection performance, resulting in a more robust and reliable ability to detect fake news on social media. In this paper, we propose the use of novel features and methods that enhance the field of fake news detection. Our findings underscore the crucial role of emotions in detecting fake news and provide valuable insights into how machine-learning models can be trained to recognize these features.