Analyzing effects of emotions on fake news detection: a covid-19 case study

Official URL: https://risc01.sabanciuniv.edu/record=b3196259

The rapid dissemination of fake news represents an important threat to the accuracy of the information, particularly in considering the COVID-19 pandemic. In this dissertation, the significance of detecting fake news has been studied, with particular attention paid to the impact that sentimental and emotional characteristics can have on the process of identifying it. On a COVID-19 Twitter dataset with labeled classes, the feelings and emotions of fake news against real news are compared. Lexiconbased sentiment analysis and emotion extractions methods are utilized for extracting the sentiments and emotions of the tweets. Three different sentiment lexicons are employed to generate the matching sentiment for each tweet, and the best performing lexicon is selected using a variety of techniques. Vader sentiment lexicon provides the most effective results. According to the sentiments displayed by Vader, fake news involve larger quantity of negative emotions than positive emotions. The tweets are evaluated with the NRC emotion lexicon, which allows for the extraction of eight basic emotions, including anticipation, anger, joy, sadness, surprise, fear, trust, and disgust. It has been discovered that negative feelings like fear, anger, and disgust are more prevalent in fake news than they are in real news. These emotions are also expressed, in a more powerful manner, via fake news. On the other hand, feelings such as trust, joy, and anticipation are more prevalent in real news, both in terms of the amount of such feelings and the intensity with which they are expressed. According to the findings, feelings have the potential to play an important role as elements in the development of fake news identification models. iv The SVM, Naive Bayes, Random Forest machine learning, and BERT deep learning models are implemented in order to validate this hypothesis. Comparisons are made between the performance of the models with and without the inclusion of emotional details. The findings show that incorporating emotional aspects into fake news detection models improves the performance of the detection model. This dissertation introduces novel features and approaches that contribute to the advancement of the field of detecting fake news. The findings highlight the significant emotional and sentimental differences among fake and real news on the COVID-19 twitter data set and highlight the important role that emotions play in the detection of fake news and provide useful insights into the process of training fake news detection models to recognize and make efficient use of these features.