machine learning fake news social media Twitter user susceptibility school shootings
Abstract:
While misinformation and disinformation have always existed in society, the prevalence of fake news on social media threatens to create divisions in society and erode trust in real news. Researchers have studied characteristics of fake news and developed accurate models to identify it on social media. However, understanding the human element of this societal problem is important. This thesis studies Twitter replies to fake news posts surrounding the shooting at Marjory Stoneman Douglas High School in Parkland, Florida and proposes a model for predicting a user’s level of susceptibility to fake news by utilizing features derived from a user’s friendship network. The features include user-based, clustering, centrality, degree, and psychology-based features. The final model, gradient-boosted trees (XGBoost) trained on a combination of 27 features from the aforementioned feature categories, achieved an AUC of 0.715. This model can be used in tandem with existing fake news detection models to create a sliding-scale intervention method based on predicted user susceptibility.