Distributed learning-enabled federated large language model for fake news detection

Abstract

Fake news detection is the process of detecting misinformative or crafted content spread through social media. Fake news causes social instabilities, increases the data leakage risks, and affects the user's privacy. The prevailing fake news detection models failed to tackle certain challenges, including overfitting issues, reduced generalizability, and computational complexity. Hence, the research proposes the Biometric Authenticated-Distributed learning enabled federated large language model (Bio-DFLLM) to detect fake news. The utilization of the hybrid biometrics enhances multiple users' privacy by sharing data with the aggregator center. The introduction of multiple feature extraction techniques learns intricate features and simplifies the detection process by lowering the computational demands and mitigating overfitting challenges. Furthermore, the pretrained response fact checker strengthens the research reliability using its domain adaptation and reliability reinforcement characteristics. In essence, the TensorFlow mirrored strategy-based distributed learning and Federated learning reduce the computational issues and enable better scalability using their data parallelism and model aggregation characteristics. Experimental results prove the superiority of the fake news detection model by achieving 98.35% accuracy, 98.40% sensitivity, 98.27% precision, and 98.40% recall using the dataset named fake news detection, for 90% training.