Leveraging GNNs and Node Entropy for Anomaly Detection: Revealing Misinformation Spreader on Twitter Network

Abstract

The rapid growth of social media, misinformation propagation has become a critical challenge, especially on platforms like Facebook and Twitter. Detecting misinformation spreaders is vital to mitigate its harmful impact on users and society. This paper proposes an innovative approach to identify potential anomalous nodes of misinformation spreaders on Twitter networks by employing Graph Neural Networks (GNNs) and entropy-based method. Utilizing GNNs, we learn node embeddings that capture the intricate patterns of information diffusion and user attributes. Additionally, we analyze the entropy of node attributes on the embeddings to identify nodes exhibiting attribute distributions significantly deviating from the normal. Those anomalous nodes exhibit in the class of misinformation spreader will lead to detect potential of aggressive node in spreading further misinformation.

Through extensive experiments conducted on real-world Twitter datasets containing misinformation-related content, our novel approach showcases its efficacy in identifying potential anomalous nodes as misinformation spreaders across various categories. By harnessing the capabilities of Graph Neural Networks (GNNs) and integrating them with entropy-based techniques via node embeddings, our methodology offers a promising avenue for gaining deeper insights into the behavior of distinct misinformation spreaders and their potential influence on others.