A Supervised Learning Ensemble for Exposing Fake News on Twitter
Fake news are a hazardous instrument for manipulation. In online social networks, fake messages and rumors are spread, usually with the intention of deceiving users and manifesting certain opinions. To preserve integrity of media, technologies are sought to expose fake news. This thesis attempts to differ fake news and real news. For this purpose, methods of natural language processing are applied. A supervised approach combining various text representations and using an ensemble of multiple classification models such as SVM, Artificial Neural Networks, XGBoost, Random Forest and AdaBoost is proposed. Finally, the approach is evaluated using the dataset of the PAN task "Profiling Fake News Spreaders on Twitter" and the results are discussed.