Anger or love? emotions as complementary signals beyond text for fake news detection

While access to the latest news through social media platforms is becoming increasingly convenient, exposure to false or incorrect information, also referred to as fake news, is also inevitable. Most efforts to identify fake news in the literature have focused on the content of the post, overlooking the textual and emotional information that may be hidden in user comments. In this paper, we investigate the effectiveness of using user comments and their emotional signals in identifying fake news. We adopt a graph-based framework that combines textual, visual, and emotional information from posts and user comments to automatically detect fake news. Our results, based on a real-world dataset collected from the Reddit platform, consisting of roughly 300,000 posts and 5 million associated user comments, suggest that using just a few user comments and their associated emotions can significantly improve the performance of a fake news detection model. To further evaluate our model on a different platform, X, we use a publicly available PHEME dataset. In addition, we conduct an emotional analysis of different categories of fake news, examining them from both the publisher and user perspectives. Interestingly, we find that the nature of fake news varies according to the publisher's underlying motivations. Some content is designed to generate controversy or polarity among audiences, while others are crafted to increase user engagement. This led to the creation of content with different tones, from negative emotions such as anger or sadness to emotions such as love or care. We further show that different fake news narratives elicit distinct emotional responses. Some evoke amusement, love, or approval from the audience, while others provoke strong disapproval, surprise, or frustration.