Existing methods for detecting and analyzing fake news often treat the problem as either a natural language processing task or a network propagation problem. This paper introduces BigSEIR, a hybrid framework combining BigBird, a transformer-based natural language processing model, with complex spreading epidemic model SEIZ (Susceptible-Exposed-Infected-Sceptic) to analyze and predict the temporal spread of fake news on Twitter. Temporal networks, representing dynamic interactions between users, are integrated into the model to capture the evolving nature of information propagation. Our approach provides a scalable and accurate method for understanding and mitigation of the spread of fake news on temporal networks by utilizing the dynamic modeling capabilities of epidemic frameworks and the semantic capabilities of BigBird for content analysis.