Hybrid Explainable Fake News Detection Using CNNs and Frequent Pattern Mining

Abstract

For smart societies and digital communication ecosystems, the wide spread of fake news on online platforms is a big challenge. This highlights the need for reliable and explainable detection methods. Deep learning approaches often achieve strong predictive performance; however, they usually work as black boxes and lack explainability. In this paper, we propose a hybrid approach for fake news detection that integrates a Convolutional Neural Network (CNN) text classifier with frequent pattern mining to generate human-understandable explanations. The proposed CNN model performs a binary classification of news content into real or fake news, while the pattern mining approach allows identifying discriminative frequent n-gram patterns for each class and links them to the label predicted by the CNN model to justify predictions. Experimental evaluations were conducted on a fake news dataset and the results show that the proposed approach scores high prediction performance while providing clear and evidence-based explanations.