Answers are grounded in this paper and cite their sources.
The rapid proliferation of fake news on social media has emerged as a significant threat to information credibility and public trust. Fake news detection approaches can be broadly categorized into two main groups: text-based and structure-based methods.Detecting fake news solely based on textual content is often insufficient, because fake news often exhibits linguistic patterns and writing styles similar to those of genuine news.Moreover, the scarcity of datasets capturing hierarchical information propagation patterns has hindered progress in low-resource languages such as Persian. To address these limitations, we introduce P-HNet, a novel Persian fake news dataset collected from X (formerly Twitter) that captures information propagation at both macro and micro levels. The macro level represents how news spreads from original publishers to users who repost it, while the micro level captures conversations and interactions surrounding the news, including user replies, community responses, and engagement patterns. Building on this hierarchical representation, we propose PGT, a hybrid feature-fusion framework that combines features extracted from news content and propagation networks. Experimental results on P-HNet demonstrate the effectiveness of PGT, achieving an F1-score of 74.51% and an accuracy of 82.19%, outperforming baseline methods such as DSS, HPFN, and conventional machine-learning models. Ablation studies further highlight the importance of textual representations, micro-level propagation patterns, and emotion-related features in detecting Persian fake news.
No takes yet. Share an insight, caveat, or question.
Hosseinbeigi et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: