Unfolding the Misinformation Spread: An In-Depth Analysis Through Explainable Link Predictions and Data Mining

In our interconnected world, the dissemination of misinformation has emerged as a crucial and pressing challenge. Social media platforms and technological advancements facilitate the proliferation of false information, thereby leading to significant repercussions on societal, political, and economic fronts. Recent research suggests using Graph Neural Networks (GNNs) to represent relationships among network actors and consequent prediction activities like pinpointing influential nodes and detecting communities. This work exploits a GNN to make link predictions on a graph representing information about misinformation tweets, their authors, and their spread. The objective is to comprehensively investigate the specific attributes of online pathways that compel users to share and amplify inaccurate information. In this sense, starting from an existing dataset of misinformation tweets, the proposed approach first applies an explainability method to each prediction, then, through frequent itemset mining, tries to detect patterns among collected explanations. Results of qualitative and quantitative research questions mainly demonstrate the contribution of interpersonal aspects to misinformation tweets spreading. To the best of our knowledge, this is the first approach exploiting a combination of Explainable Artificial Intelligence (xAI) and Data Mining to GNNs in fake news spreading analysis for prevention and mitigation purposes.