An Adaptive and Unsupervised Graph-based Approach for Fake News Detection

Detecting fake news has become a significant challenge in the digital era today, where false information spreads rapidly across various platforms. Current methods for detecting fake news often overlook the social context of how the news spreads and rely primarily on labeled datasets. As a result, they often struggle to effectively capture the nuanced relationships between news topics, entities, and content. This paper aims to address these challenges by proposing an unsupervised and adaptive framework that utilizes a variant of the Graph Sage Autoencoder to process heterogeneous graph data for improved fake news detection. Our approach captures complex relationships among articles, entities, and topics using adaptive data augmentation techniques such as node feature masking and edge dropping, enhancing robustness and generalization. Using SAGEConv layers within a contrastive learning framework, it produces informative embeddings for node classification, incorporating adaptive feature masking and edge dropping to selectively balance and refine the graph structure. Experimental results highlight the superiority of our approach compared to state-of-the-art algorithms, demonstrating improved detection accuracy and robustness while ensuring high scalability for fake news detection in complex information networks.