EmoKnow: Emotion- and Knowledge-Oriented Model for COVID-19 Fake News Detection.

Content-based methods are inadequate for detecting fake news related to COVID-19 due to the complexity of this domain. Some studies integrate the social context information of the news to improve performance. However, such information is not consistently available and sometimes not helpful regarding COVID-19, as most users lack professional knowledge about it and may be unable to respond accurately. Additionally, fake news often employs emotional manipulation to exploit people’s emotions to shape their beliefs and actions. Therefore, we propose EmoKnow, an emotion- and knowledge-oriented model, for detecting fake news about COVID-19. Our proposed method incorporates language modeling, emotion feature extraction, and external knowledge sources to provide an informative representation of news. Experimental results on four COVID-19-related datasets show that EmoKnow significantly outperforms state-of-the-art approaches.