Trust-aware and explainable AI framework for misinformation diffusion in social media networks under uncertainty

Information diffusion on social media platforms, adopted widely across universities, educational systems, high-tech industries, governmental institutions, and by individuals, presents significant challenges due to the complex and dynamic nature of user interactions. In these contexts, misinformation (coexistence of truthful and false content) has emerged as a critical issue that existing analytical frameworks are unable to address. This study provides a trust-aware and explainable artificial intelligence (AI) approach to the dual character of online information by using generalized bipolar fuzzy relations and hyper-relational structures to model social networks and analyze the spread of misinformation in uncertain conditions. A theoretical framework is developed by integrating bipolar fuzzy logic with a graph-based approach to capture the co-occurrence of true and false information within shared content. By utilizing a content-based trust score and an engagement-normalized influence metric, the model measures users’ credibility and influence within the network. The framework introduces a truth membership value in the interval [0,1] and a falsehood membership value in the interval (Formula presented) to detect the simultaneous representation of positive and negative information. The model further defines a trust score T_u to effectively evaluate user reliability and an influence degree based on normalized engagement metrics. For large-scale analysis, community detection techniques are employed to examine how misinformation spreads across different social clusters and interaction layers. The proposed approach is validated using large-scale data from the Bluesky, MuMiN, and LIAR benchmark datasets on social media networks. The results show that while the majority of users display a relatively low level of activity and influence, a minority of users have a disproportionately large effect on the spread of misinformation. The results demonstrate the effectiveness of bipolar uncertainty modeling in capturing credibility and influence patterns that are mostly overlooked by the traditional methods, providing practical insights for trust-aware and explainable social media AI analysis.