A Hybrid BiLAT–SVM Framework with Uncertainty-Aware Decision Mechanism for Fake News Detection

Abstract

The rapid dissemination of misinformation on social media needs the creation of precise and efficient detection systems to maintain information integrity and public trust. This study introduces the Uncertainty- and Confidence-Adaptive BiLAT-SVM (UCA-BiLAT-SVM) model, which integrates deep learning and machine learning techniques for the classification of fake news. The system uses Support Vector Machine (SVM) for categorising the obtained latent space and employs Bidirectional Long Short-Term Memory with Attention (BiLAT) for acquiring contextual and semantic representations of words. An entropy-based uncertainty mechanism is used to adaptively select or combine predictions when cases are ambiguous, to increase the reliability of the predictions. The use of the Bayesian Optimization with Hyperband (BOHB) algorithm for hyperparameter optimization. The proposed model is tested on Fake & Real News, FakeNewsNet and ISOT datasets with the commonly used performance metrics. Experimental findings show good performance results with accuracy up to 99.5% and F1 score up to 99.4%. The results show the effectiveness and scalability of the proposed approach for fake news identification that combines contextual representation learning, margin-based classification and uncertainty-aware decision-making.