Unmasking Rumors on Social Media by Combining Affective Patterns Flow with User Behaviors

Social media, a platform for rapid information dissemination, has undergone a significant shift. Its users are no longer passive recipients but active creators of content. This paradigm shift has led to a surge in the quantity and diversity of information, reducing their credibility and paving the way for the production and spread of rumors. The prevalence of rumors poses a severe threat to the trustworthiness of the entire social media ecosystem. The adverse impact of rumors underscores the urgent need for their detection, making it a pressing concern and a crucial requirement. This paper introduces a rumor detection approach that combines affective and psycholinguistic information flow on social media, user attributes, propagation structures, and a word embedding model. It leverages external lexicon-based knowledge to extract the affective and psycholinguistic information flow on social media. It entails detecting rumors utilizing users’ profiles and propagation structure information, and uses word-level GloVe embedding to preserve the semantic relations of the posts and their underlying affective information. The proposed approach is evaluated over the publicly available ${\text{Twitter}}/\mathbb{X}$ dataset and compared to eight baselines and state-of-the-art approaches, demonstrating substantially remarkable performance for rumor detection on social media.