Presenter: Mouna Rabhi, PhD Candidate
Date: May 31, 2022
Fake news is a serious concern that has received a lot of attention lately due to its harmful impact on society. In order to limit the spread of fake news, researchers have proposed automated ways to identify fake news articles using artificial intelligence and neural network models. However, existing methods do not achieve a high level of accuracy, which hinders their efficacy in real life. To this end, we introduce FN2 (Fake News detectioN): a novel neural-network-based framework that combines both textual and contextual features of the news articles. Among the many unique features of FN2, it utilizes a set of explicit contextual features that are easy to collect and already available in the raw user metadata. To evaluate the accuracy of our classification model, we collected a real dataset from a fact-checking website, comprising over 16 thousand politics-related news articles. Our experimental results show that FN2 improves the accuracy by at least 13%, compared to current state-of-the-art approaches. Moreover, it achieves better classification results than the existing models, even when relying only on the textual data from the news articles. Finally, preliminary results also show that FN2 provides a quite good generalization—outperforming competitors—also when applied to a qualitatively different data-set (entertainment news). The novelty of the approach, the staggering quantitative results, its versatility, as well as the discussed open research issues, have a high potential to open up novel research directions in the field.