Misinformation datasets

For this study, we used three publicly available, widely used datasets for the misinformation detection task. The benchmark misinformation datasets: (1) FakeNewsNet: 405K news articles from PolitiFact (142K real, 263K fake) and Gossipcop (246K real, 159K fake), spanning 2016-2020; includes article content, social engagement networks, and user profiles. (2) LIAR-PLUS: 12.8K claims from PolitiFact with fine-grained labels (pants on fire, false, barely true, half-true, mostly true, true), evidence sentences, and speaker metadata; extends the original LIAR dataset. (3) Twitter Rumour Datasets (PHEME): 4.9K rumour threads collected during 2011-2016 major events (Ferguson, German election, Sydney siege), with structural information (retweet\/quote graphs) and fine-grained stances (support, deny, query, comment)