Combining transformer, Bi-LSTM and CNN for efficient Vietnamese fake news detection

Abstract

Fake news spreading widely on digital platforms presents challenges where the rapidly evolving information landscape lacks effective detection. This paper introduces a hybrid deep learning model that integrates Transformer, Bidirectional Long Short-Term Memory (Bi-LSTM), and Convolutional Neural Network architectures to jointly capture both semantic and structural characteristics of Vietnamese text for the task of fake news detection. A high-quality dataset comprising 2,336 manually annotated Vietnamese news articles was developed, spanning three major domains: Politics, Healthcare, and Society. The proposed model was trained and evaluated on this dataset and benchmarked against two widely adopted baseline models: Gated Recurrent Unit and Bi-LSTM. Experimental results indicate that the proposed model achieves superior performance, attaining an overall accuracy of 95.3% and an F1-score of 0.951. These findings underscore the efficacy of combining multi-layered linguistic representations in enhancing Vietnamese fake news detection and contribute a valuable annotated resource for future studies in this domain.