For real-world rumor detection: Anomaly detection framework using graph teacher control learning

Total: 1

Anecdotal detection methods, based on learning about communication structures, are being used primarily as a classification task for limited label data. However, social media data from the real world show that the number and distribution of rumours in a large number of official publications is low and uneven. In order to address data scarcity and imbalances, we have constructed two large-scale data sets for dialogue from and analysis of domain name distributions from Weibo and Weibo. We have found a clear difference between the spread of rumours and non-fiction, which are mostly in the area of entertainment, and the concentration of rumours in the field of journalism, which suggests that the detection of rumours is consistent with the anomaly detection paradigm. It treats unlabelled data as non-rumers in a high-pressure manner and adjusts graphic contrasting learning for anecdotal testing. Extensive experiments have shown that AD-GSCL is superior to class balance, imbalance, ejection, etc. Our research has provided valuable insight into real-world rumour detection, characterized by uneven data distribution.