ZoKa: a fake news detection method using edge-weighted graph attention network with transfer models

Recent advances in social networks enable users to communicate and share their ideas. While social networks are beneficial to our society, it can lead to exponential disinformation growth. The extensive spread of such unreliable information and fake news can lead an adverse impact on the public opinion and cause uncertain outcomes of public events. Despite the recent progress in identifying fake news, it is still a time-consuming, complex, and diverse task. To address these challenges, we first generate a user friendship and retweet propagation graph to filter the potential fake users by leveraging Graph Attention Network. The underlying user graph is built by the functional connectivity matrices and the node features including both the user–user connections regarding their activities on the Twitter social network. Also, we generate a content graph including comments, profile descriptions of potential fake users along with their shared news contents. Then, we employ a novel fake news detection method on the generated content graph based on the Edge-weighted Graph Attention Network) using pre-trained encoders. The results from the experiments conducted on two real-world datasets show that our method achieves remarkable results when compared to the existing approaches in terms of Accuracy and F1 scores.