An Adaptive and Energy-Efficient Edge Framework for Real-Time Fake News Detection

Automated fact-checking in newsrooms demands high accuracy and real-time performance on resource-constrained devices. We propose a conceptual framework that integrates meta-learning, split learning, and EdgeBERT-based inference for efficient, real-time misinformation detection. The framework filters redundant claims via a local knowledge base, employs an early exit mechanism in EdgeBERT to reduce latency and energy consumption, and uses split learning to distribute computational tasks between the edge and the cloud. A meta-learning component continuously adapts the model to emerging misinformation patterns. Our approach offers a scalable and energy-efficient solution for automated fact-checking in environments such as newsrooms and mobile applications, with future work aimed at fine-tuning model components and validating performance under real-world conditions.