This paper addresses the increasing spread of misinformation and disinformation on social networks by proposing a novel behavioral forensics model. The model categorizes users into five classes based on their actions after exposure to both misinformation and its refutation, aiming to understand user intent. Network features, extracted using graph embedding models (LINE and PyTorch-BigGraph), are combined with user profile features to train a machine learning model. Evaluated on a Twitter dataset, the model achieves high precision and recall in detecting malicious actors, significantly outperforming baseline models.
